[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"navigation-en":3,"articles-en":24,"header-me":37,"posts-en":38},[4,8,12,16,20],{"title":5,"path":6,"stem":7},"The Humanitarian AI Chatbot","\u002Fprojects\u002Fai-chatbot-project","1.projects\u002F1.ai-chatbot-project",{"title":9,"path":10,"stem":11},"Analytics Dashboard for ReliefWeb Jobs Data","\u002Fprojects\u002Freliefweb-jobs-dashboard","1.projects\u002F2.reliefweb-jobs-dashboard",{"title":13,"path":14,"stem":15},"Understanding Cash Assistance in Ukraine","\u002Fprojects\u002Fhumanitarian-data-platform","1.projects\u002F3.humanitarian-data-platform",{"title":17,"path":18,"stem":19},"Humanitarian Data Analyst Portfolio","\u002Fprojects\u002Fweb-development-project","1.projects\u002F4.web-development-project",{"title":21,"path":22,"stem":23},"AI-Powered Humanitarian Legal Advisor for Ukraine","\u002Fprojects\u002Fhumanitarian-paralegal-advisor","1.projects\u002F5.humanitarian-paralegal-advisor",{"id":25,"title":26,"body":27,"description":28,"extension":29,"meta":30,"navigation":31,"path":33,"seo":34,"stem":35,"__hash__":36},"articles\u002F3.articles.yml","Articles",null,"Stay updated with our latest articles and insights","yml",{},{"icon":32},"i-lucide-newspaper","\u002Farticles",{"title":26,"description":28},"3.articles","CwXqu_Sg9j5sS44ytTsQ2Z2HDAY0W0eZIem7V6tMs38",{"user":27,"profile":27},[39,274,593,1267,1411,1759,2036,2349,2428,2689,2999],{"id":40,"title":41,"authors":42,"badge":48,"body":50,"date":264,"description":265,"extension":266,"external_links":27,"image":267,"meta":268,"navigation":269,"path":270,"seo":271,"stem":272,"__hash__":273},"posts\u002F3.articles\u002F1.presentation.md","A New Portfolio of AI and Data Solutions for the Humanitarian Sector",[43],{"name":44,"to":45,"avatar":46},"Jesus Baena","https:\u002F\u002Flinkedin.com\u002Fin\u002Fjbaenanet",{"src":47},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Favatar-jesus-color.jpg",{"label":49},"AI | DataAnalysis",{"type":51,"value":52,"toc":247},"minimark",[53,57,61,69,98,105,115,162,166,171,178,181,184,188,191,197,205,209,212,216,219,223,226,230,233,237,240,244],[54,55,41],"h2",{"id":56},"a-new-portfolio-of-ai-and-data-solutions-for-the-humanitarian-sector",[58,59,60],"p",{},"In the humanitarian sector, our most valuable lessons are learned on the ground. Yet, these insights often remain isolated—locked in field offices, buried in reports, or confined to individual experience. Restrictive agency policies can discourage sharing, leaving many of us to reinvent tools that we know must already exist.",[58,62,63,64,68],{},"I have worked over fourteen years on short-term rapid onset emergencies such as Haiti, Niger or The Philippines as in long-term deployments in the DRC, South Sudan, and now Ukraine, I've seen this pattern firsthand. I believe we have too much to share to let that expertise stay siloed. That’s why I’ve launched my own web site last year. Today, I am upgrading significatly this platform: ",[65,66,67],"strong",{},"baena.ai","— with the intend of contributing to the transformation of field challenges into practical, technology-driven solutions.",[70,71,77,89],"div",{"className":72},[73,74,75,76],"grid","grid-cols-2","gap-4","my-8",[70,78,79],{},[58,80,81],{},[82,83],"img",{"alt":84,"className":85,"src":88},"PortAuPrince",[86,87],"rounded-lg","w-full","https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Ftacloban.png",[70,90,91],{},[58,92,93],{},[82,94],{"alt":95,"className":96,"src":97},"Tacloban",[86,87],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fhaiyan.png",[58,99,100,101,104],{},"Rather than merely a list of projects. It’s a curated showcase of my work as a ",[65,102,103],{},"Humanitarian Data Analyst",", demonstrating how I design and build custom AI, data analysis, and automation solutions tailored to the unique demands of aid work.",[58,106,107,108,110,111,114],{},"For those interested in the journey behind the solutions, I've created a separate space to share my process. While ",[65,109,67],{}," showcases the finished, impactful products, my personal digital garden at ",[65,112,113],{},"baena.blog"," offers a look behind the curtain. It's where you can follow the building process for these tools, explore my thoughts-in-progress, and see the experiments that lead to field-ready technology.",[70,116,121,135,144,153],{"className":117},[73,118,119,120,76],"grid-cols-1","md:grid-cols-2","gap-2",[122,123,132],"card",{"className":124,"icon":128,"target":129,"title":130,"to":131},[125,126,127],"text-xs","px-3","py-1.5","i-heroicons-chart-bar","_blank","Relief Jobs by Experience","https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FEvolution-of-relief-jobs-by-experience-levels",[58,133,134],{},"Analysis of job opportunities across experience levels.",[122,136,141],{"className":137,"icon":138,"target":129,"title":139,"to":140},[125,126,127],"i-heroicons-truck","Fleet Management Compliance","https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FField-Management-Kit\u002F3.3.1.4.-Donor-Compliance-Essentials-for-Fleet-Management",[58,142,143],{},"Donor compliance guidance for vehicle operations.",[122,145,150],{"className":146,"icon":147,"target":129,"title":148,"to":149},[125,126,127],"i-heroicons-document-text","Obsidian Knowledge Management","https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FObsidian",[58,151,152],{},"Personal knowledge management for humanitarian work.",[122,154,159],{"className":155,"icon":156,"target":129,"title":157,"to":158},[125,126,127],"i-heroicons-server-stack","Git-Based Backup System","https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FA-Personal-Git-Based-Backup--and--Replication-System-for-my-new-Omarchy-setup",[58,160,161],{},"Custom backup and replication architecture setup.",[54,163,165],{"id":164},"the-humanitarian-assistant","The Humanitarian Assistant",[167,168,170],"h3",{"id":169},"an-ai-powered-copilot-for-humanitarian-data-analysis-and-decision-making","An AI-Powered Copilot for Humanitarian Data Analysis and Decision-Making",[58,172,173,174,177],{},"The centerpiece of this launch is the upgraded ",[65,175,176],{},"Humanitarian Agent",". After attracting users from across the sector and continents, its underlying technology has been modernized while preserving the powerful knowledge base that makes it so effective.",[58,179,180],{},"Specifically, there are huge technological improvements under the hood: a new stack based on the principles of resilience, ownership, and privacy. The Humanitarian Agent now runs on a custom-built infrastructure that I control, ensuring data sovereignty and compliance with humanitarian data protection standards.",[58,182,183],{},"Additionally, new capabilities have been added to enhance its utility. The agent can now (aside of checking on its own knowledge base): do searches to solid sources, access and analyze exernal databases, such as ACAPS, ReliefWeb, or OCHA Financial Tracking Service, and connect to bespoke APIs and data sources that humanitarian organizations use in the field.",[54,185,187],{"id":186},"other-demos","Other demos",[58,189,190],{},"Aside from the humanitarian agent project, I have been working on several other AI and data analysis solutions for the humanitarian sector. Some of these include: a real-time dashboard for monitoring the jobs market in the humanitarian sector. A Web Map Server (WMS) for visualizing geospatial data related to humanitarian situation in Ukraine. And a data pipeline for automating the collection and analysis of humanitarian data from multiple sources.",[58,192,193,194,196],{},"These projects, along with the Humanitarian Agent, are now featured on ",[65,195,67],{},". Each solution is designed to address specific challenges faced by humanitarian workers, leveraging AI and data analysis to improve efficiency and decision-making in the field. The most fluid section is the one regarding static data analysis and visualization solutions, where I will be adding new projects on a regular basis.",[198,199],"nuxt-img",{"alt":200,"className":201,"src":204},"Analytics schema",[86,202,203,76],"w-3\u002F4","mx-auto","https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fimage7.png",[54,206,208],{"id":207},"what-other-content-is-coming","What other content is coming?",[58,210,211],{},"I’m rolling out a steady stream of new resources that will help teams turn raw data into rapid, evidence‑based action, they are tools I am rescuying from my hard drives, and freshing up for wide dissemination. Roughly every month, I will be publishing a new demo or article that dives into a specific technology or use case relevant to humanitarian operations.",[167,213,215],{"id":214},"ai-and-business-intelligence-solutions","AI and Business Intelligence Solutions",[58,217,218],{},"Move beyond static reports to a dynamic decision-support environment. I build custom AI Agents that act as expert advisors for your team and interactive Business Intelligence dashboards that provide real-time operational clarity. This service transforms your raw data into a strategic asset, enabling your organization to make faster, more evidence-based decisions when it matters most.",[167,220,222],{"id":221},"data-analysis-and-impact-reporting","Data Analysis and Impact Reporting",[58,224,225],{},"Fulfilling donor requirements and proving programmatic impact shouldn't be a bottleneck. I design robust data analysis pipelines that automate the entire process, from cleaning messy field data to generating compelling, visualized reports. This service provides you with the irrefutable evidence you need for accountability and saves your Monitoring & Evaluation team hundreds of valuable hours.",[167,227,229],{"id":228},"business-process-and-workflow-automation","Business Process and Workflow Automation",[58,231,232],{},"Free your team from the drudgery of manual data entry and repetitive administrative tasks. I audit your key operational processes, identify bottlenecks, and build powerful automation systems that connect your existing tools into a seamless flow. This service gives your team back its most valuable resource—time—while reducing errors and dramatically improving overall efficiency.",[167,234,236],{"id":235},"custom-web-application-development","Custom Web Application Development",[58,238,239],{},"When off-the-shelf software doesn't meet your unique operational needs, a custom solution is the answer. I provide end-to-end web application development, from initial concept to deployment and long-term support. With optional managed care, I ensure your bespoke tool remains secure and effective, allowing you to own a digital asset that is perfectly tailored to your workflow without the burden of ongoing maintenance.",[167,241,243],{"id":242},"sector-specific-capacity-building","Sector-Specific Capacity Building",[58,245,246],{},"Technology is only effective when people are empowered to use it. Drawing on my experience as a humanitarian leader, I design and deliver customized capacity-building programs—from hands-on workshops to one-on-one mentoring. This service ensures your team has the skills and confidence to embrace new tools, guaranteeing the long-term sustainability and impact of your digital transformation efforts.",{"title":248,"searchDepth":249,"depth":249,"links":250},"",2,[251,252,256,257],{"id":56,"depth":249,"text":41},{"id":164,"depth":249,"text":165,"children":253},[254],{"id":169,"depth":255,"text":170},3,{"id":186,"depth":249,"text":187},{"id":207,"depth":249,"text":208,"children":258},[259,260,261,262,263],{"id":214,"depth":255,"text":215},{"id":221,"depth":255,"text":222},{"id":228,"depth":255,"text":229},{"id":235,"depth":255,"text":236},{"id":242,"depth":255,"text":243},"2023-08-25T00:00:00.000Z","Bridging 14 Years of Field Experience with Practical, Field-Ready Technology","md",{"src":88},{},true,"\u002Farticles\u002Fpresentation",{"title":41,"description":265},"3.articles\u002F1.presentation","z-hLB1w8YfX_UPQe8MTH_lbPmKU1I3Q_wcXl4zlcc7Q",{"id":275,"title":276,"authors":277,"badge":280,"body":282,"date":577,"description":578,"extension":266,"external_links":579,"image":585,"meta":588,"navigation":269,"path":589,"seo":590,"stem":591,"__hash__":592},"posts\u002F3.articles\u002F10.data-sovereignty-ngo.md","Do No Harm Starts With Where You Keep the Data",[278],{"name":44,"to":45,"avatar":279},{"src":47},{"label":281},"Data Sovereignty | Governance",{"type":51,"value":283,"toc":567},[284,287,296,299,307,310,314,321,330,333,336,340,343,346,349,356,359,362,365,369,377,380,383,386,389,393,396,399,405,411,417,420,424,427,433,447,450,454,460,463,469,476,480,483,486,489,499,502,516,519,522,526,538,541,544,551,554,559],[58,285,286],{},"In June 2025, a senior director from Microsoft France was asked under oath, in front of the French Senate, whether he could guarantee that French public data hosted in Microsoft's French datacentres would never be handed to US authorities. He could not. He explained that Microsoft challenges such requests where it can, and publishes the numbers. But the guarantee does not exist.",[58,288,289],{},[290,291,295],"a",{"href":292,"rel":293},"https:\u002F\u002Fwww.senat.fr\u002Fcompte-rendu-commissions\u002F20250609\u002Fce_commande_publique.html",[294],"nofollow","The transcript is public.",[58,297,298],{},"That was the day a comfortable assumption died. Many of us had been operating on the hopeful presumption that data residency was enough — that if the servers were in Frankfurt or Paris, our beneficiaries were covered and our operations were safe. Microsoft and Google both offer EU hosting and both take GDPR compliance seriously. That part is real.",[300,301,304],"insight-box",{"icon":302,"title":303},"i-ph-scales-duotone","Geography vs. jurisdiction",[58,305,306],{},"Residency is about geography. Sovereignty is about jurisdiction. The CLOUD Act of 2018 reaches data controlled by US-incorporated companies regardless of where it physically sits. You do not need to invoke the Patriot Act or anything exotic. The ordinary law is sufficient, and the company itself has now said so in a European parliament.",[58,308,309],{},"Nothing against US services in particular. I use several of them myself and some are excellent. But we work in a sector with a first principle, and the principle is Do No Harm. In 2026, Do No Harm starts with where you keep the data.",[54,311,313],{"id":312},"two-risks-not-one","Two risks, not one",[58,315,316,317,320],{},"When we talk about this, we usually mean confidentiality — who can read our beneficiary lists. That is real. But there is a second risk that gets much less attention and hits harder in the field: ",[65,318,319],{},"continuity",".",[58,322,323,324,329],{},"US providers must comply with the Office of Foreign Assets Control (OFAC), and this is a very real concern. In August 2026, EU leaders had to publicly rally behind the International Criminal Court ",[290,325,328],{"href":326,"rel":327},"https:\u002F\u002Fwww.euronews.com\u002Fmy-europe\u002F2026\u002F08\u002F19\u002Feu-leaders-rally-behind-international-criminal-court-after-fresh-us-sanctions",[294],"after a fresh round of US sanctions",". A mandated international body, facing service exposure because of a decision taken in another jurisdiction entirely.",[58,331,332],{},"Now imagine that logic applied to a country office. One day just an email account that stops working on a Tuesday because a compliance team somewhere applied a rule you were never party to and cannot appeal. This is a more real possibility these days.",[58,334,335],{},"That is a programme risk, and it belongs on a risk register next to access negotiations and supply chain.",[54,337,339],{"id":338},"so-what-is-actually-in-your-cloud","So what is actually in your cloud?",[58,341,342],{},"Start here, before any discussion about vendors or migrations.",[58,344,345],{},"Ask the question honestly: what is in there? Not what the policy says should be in there. What is actually in there.",[58,347,348],{},"In my experience the answer is always the same and always worse than expected. Beneficiary lists in Excel on SharePoint. Case notes in Word documents on OneDrive. Protection information in email threads. Registration data in a folder someone created for a 2019 response and never closed.",[300,350,353],{"icon":351,"title":352},"i-ph-folder-lock-duotone","A category error",[58,354,355],{},"Beneficiary data and case management do not belong in a general-purpose file system. Not in SharePoint, not in OneDrive, not in Dropbox — and not in Nextcloud either, before anyone thinks this is only a Microsoft problem. It is not. It is a category error.",[58,357,358],{},"A file system cannot do what case data requires. Permissions are almost impossible to manage properly at scale. Changes are not tracked in any way that survives an audit. There is no archiving discipline, no retention that actually executes, no meaningful record of who looked at what and when. For protection data, that access history is often the part that made the holding defensible in the first place.",[58,360,361],{},"We got away with it for years because nobody could search across all of it at once.",[58,363,364],{},"That just ended.",[54,366,368],{"id":367},"copilot-did-not-create-the-problem-but-it-made-it-visible","Copilot did not create the problem, but it made it visible.",[58,370,371,372,376],{},"Copilot and Gemini surface anything the person asking has permission to see. Not what they knew they could see. What they ",[373,374,375],"em",{},"can"," see.",[58,378,379],{},"Most NGO tenants have a decade of permission sprawl. Sites shared with \"everyone in the organisation\" during a surge. Folders inherited from a closed project. Files technically readable by three hundred people and practically buried where nobody would ever look.",[58,381,382],{},"Copilot un-buries them. A programme officer asking a reasonable question can now surface a protection case file that permissions always allowed and no one ever expected them to find.",[58,384,385],{},"And the feature arrived by release note, not by procurement decision. Nobody chose this. It appeared.",[58,387,388],{},"Before enabling any AI assistant on your tenant, run a permissions audit. It is cheap, it takes days not months, and it is valuable whether or not you ever switch the AI on — because the exposure already exists. Copilot only lit it up.",[54,390,392],{"id":391},"the-shortcut-that-does-not-work","The shortcut that does not work",[58,394,395],{},"Some organisations reach for what feels like the safe answer: hyper-anonymisation. If we cannot protect the data, we will not hold the data. Strip the identifiers and sleep well.",[58,397,398],{},"I understand the instinct. It is still wrong, for three reasons.",[58,400,401,404],{},[65,402,403],{},"Anonymisation breaks."," Re-identification from combined attributes is well established and it gets easier every year. Location, age band, household composition, date of assistance — put enough weak identifiers together in a small caseload and you have a name. If your \"anonymised\" dataset leaks, or Copilot surfaces it, you are not off the hook.",[58,406,407,410],{},[65,408,409],{},"Your donors will not accept it."," Try explaining to an institutional donor that you cannot identify the people you served. Accountability to affected populations, verification, audit — all of it assumes you know who received what.",[58,412,413,416],{},[65,414,415],{},"And it destroys coordination."," Without identity you cannot deduplicate. You cannot deconflict with another agency working the same district, or with the line ministry, or with the cluster. So you get double registration in one place and gaps in another. That is not a privacy win. That is an operational failure with protection consequences, and the people who pay for it are the ones who were missed.",[58,418,419],{},"The answer is not to hold less. It is to hold it properly.",[54,421,423],{"id":422},"extraordinary-databases-for-extraordinary-cases","Extraordinary databases for extraordinary cases",[58,425,426],{},"This is a solved problem, and it was solved by our own sector.",[58,428,429,430],{},"If you run medical consultations, you need an electronic medical record — not a folder structure. If you run nutrition case management, DHIS2 exists and is the de facto standard across ministries of health. ",[65,431,432],{},"These tools are open source, field-proven at national scale, and they were built with the authentication, role separation, audit logging and archiving that caseload data actually requires.",[58,434,435,436,441,442,320],{},"I keep working demos if you want to see them rather than read about them: ",[290,437,440],{"href":438,"rel":439},"https:\u002F\u002Fdhis2.baena.info\u002F",[294],"dhis2.baena.info"," and ",[290,443,446],{"href":444,"rel":445},"https:\u002F\u002Ffemr.baena.info\u002F",[294],"femr.baena.info",[58,448,449],{},"The point is not these specific tools. The point is that purpose-built beats general-purpose, and that our sector already has purpose-built options that we routinely ignore in favour of a folder on SharePoint because it was already there.",[54,451,453],{"id":452},"two-traps-on-the-way-out","Two traps on the way out",[58,455,456,459],{},[65,457,458],{},"Trap one: adapting a commercial ERP."," I have watched organisations contract enterprise ERP implementations or invest seriously in Power Apps development. Good luck.",[58,461,462],{},"The mismatch is structural. Commercial software is built for the business economy, which counts and invoices earnings. NGOs spend money. The whole logic runs in the opposite direction — grants, budget lines, donor restrictions, eligibility, reporting against a proposal. Every one of those has to be forced onto a data model designed for revenue. It always takes more work than the estimate, and the estimate was already the reason you chose it.",[58,464,465,468],{},[65,466,467],{},"Trap two: building your own from scratch."," We have been here before. Enormous development cost, and then a dependency — an Oracle licence, a proprietary framework — that made changing course mid-project impossible. So the organisation ran forward and poured in more money, because stopping felt like admitting that years of work and investment were wasted. Outsourced Power Apps development lands in the same place, just with a shorter runway.",[58,470,471,472,475],{},"What has genuinely changed is the cost of building. ",[65,473,474],{},"AI has made custom tooling dramatically cheaper to produce."," NGO processes are more modular than they used to be. And mature open-source databases like PostgreSQL mean your data stays interoperable and portable instead of locked inside someone's licence.",[54,477,479],{"id":478},"so-should-you-leave-microsoft-and-google","So should you leave Microsoft and Google?",[58,481,482],{},"No. And I say that as someone who self-hosts almost everything.",[58,484,485],{},"For a mid-size organisation that has run on M365 or Workspace for a decade, migrating the whole estate is not realistic and not a good use of your political capital. Email, documents, finance admin, HR — leave them. Take the EU data boundary options, document what remains exposed honestly, and move on.",[58,487,488],{},"But do not leave it as it is either. Split the problem:",[490,491,492],"ul",{},[493,494,495,498],"li",{},[65,496,497],{},"Office tools stay. Caseload data moves."," That is the line. General organisational data can live with your current provider. Beneficiary and case data goes to purpose-built systems on infrastructure you control, in a jurisdiction you chose deliberately.",[58,500,501],{},"Two things follow from that, and both are about continuity rather than confidentiality.",[490,503,504,510],{},[493,505,506,509],{},[65,507,508],{},"Have a plan B, and write it down."," If the tenant became unavailable organisation-wide tomorrow, what happens? Who holds a contact list that is not in Outlook? Which country office can still authorise a funds transfer? Most organisations have never asked the question. It takes an afternoon, it costs nothing, and it is the cheapest business continuity work available to you.",[493,511,512,515],{},[65,513,514],{},"And keep a copy under the mattress."," Not everything — but the archive, the registration records, the closed-project files. The cloud should be redundancy, not the only copy.",[58,517,518],{},"In March 2026, Nine PBS, the public broadcaster in St. Louis, lost access to roughly 50 terabytes of material covering seventy years of local television history. Their cloud storage vendor went quiet, then defunct. In this case, it was not one of the hyperscalers but a small specialist provider they had renewed with annually since 2019. Fifty terabytes fits on a commercial NAS for less than the price of a laptop. That is the whole lesson.",[58,520,521],{},"Each organization operates differently, but the underlying principles apply universally. The world is evolving, and while a unified Humanitarian Cloud Service may eventually be the solution, its development does not excuse anyone from their immediate obligations.",[54,523,525],{"id":524},"there-is-a-clock-on-the-rest-of-it","There is a clock on the rest of it",[58,527,528,529,532,533,320],{},"The Microsoft grants already changed once. On 1 July 2025 the free Business Premium and Office 365 E1 grants for nonprofits ended, replaced by 300 free Business Basic seats and discounts of up to 75%. Organisations that waited until their renewal date got hurt. The lesson is not about pricing — it is that your entire operational stack was resting on a donation, not a contract, and donations are modified unilaterally. This transition is not going exactly smoothly: ",[65,530,531],{},"over 170,000 nonprofits have been affected",", and Slate has asked whether ",[290,534,537],{"href":535,"rel":536},"https:\u002F\u002Fslate.com\u002Ftechnology\u002F2026\u002F08\u002Fmicrosoft-software-nonprofit-data-delete.html",[294],"Microsoft is to blame for the data loss",[58,539,540],{},"Microsoft's recent licensing changes retire the free Business Premium tier, forcing NGOs to either pay for desktop applications and advanced security or downgrade to a web-only free tier that severely hinders offline field operations. Furthermore, Microsoft now strictly enforces an 85% active usage rule, where failing to maintain consistent account activity can trigger the revocation of an organization's entire donated license allocation. When these licenses are downgraded or revoked, the associated data is permanently purged within 30 to 90 days, making independent, self-hosted backup infrastructure essential to protect institutional memory.",[58,542,543],{},"Meanwhile the EU Data Act removes cloud switching charges entirely from 12 January 2027. The contract you sign this quarter will still be running on that date. If it bakes in old-world egress terms, multi-year lock-ins and auto-renewals, the deadline will arrive and find you still bound by the economics it was meant to dismantle.",[300,545,548],{"icon":546,"title":547},"i-ph-list-checks-duotone","The four things to do this quarter",[58,549,550],{},"Audit your permissions, move the caseload data, write the lockout plan, and read your contract before renewal.",[58,552,553],{},"I am writing a separate piece on AI sovereignty for NGOs. That one is a harder problem, and the good news is that it is also the layer where you still have room to move. Stay tuned.",[58,555,556],{},[373,557,558],{},"I work with humanitarian organisations on data systems, sovereign infrastructure and applied AI. If you are looking at this and not sure where to start, the permissions audit is the first move and you can do it yourself.",[560,561],"closing-cta",{"description":562,"icon":563,"link-label":564,"link-url":565,"title":566},"The permissions audit is the first move, and you can do it yourself. If you want a second pair of eyes on it, or on splitting office tools from caseload data, let's talk.","i-ph-chats-circle-duotone","Let's talk","\u002Fcontact","Not sure where to start?",{"title":248,"searchDepth":249,"depth":249,"links":568},[569,570,571,572,573,574,575,576],{"id":312,"depth":249,"text":313},{"id":338,"depth":249,"text":339},{"id":367,"depth":249,"text":368},{"id":391,"depth":249,"text":392},{"id":422,"depth":249,"text":423},{"id":452,"depth":249,"text":453},{"id":478,"depth":249,"text":479},{"id":524,"depth":249,"text":525},"2026-08-31T00:00:00.000Z","NGOs, sovereignty and the cloud. This is not a drill.",[580,583],{"label":581,"url":438,"icon":582},"DHIS2 Demo","i-ph-database-duotone",{"label":584,"url":444,"icon":582},"FEMR Demo",{"src":586,"alt":587},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00043_.png","Illustration of two people reaching through a cloud between rows of server racks to exchange documents, a laptop and a key",{},"\u002Farticles\u002Fdata-sovereignty-ngo",{"title":276,"description":578},"3.articles\u002F10.data-sovereignty-ngo","nIgloOg_Yu2LQ6tVL9TSTp7Smsd1pZ00aN4A6yIefQ8",{"id":594,"title":595,"authors":596,"badge":599,"body":601,"date":1257,"description":1258,"extension":266,"external_links":27,"image":1259,"meta":1262,"navigation":269,"path":1263,"seo":1264,"stem":1265,"__hash__":1266},"posts\u002F3.articles\u002F11.humanitarian-flagship-reports.md","The Cost of Humanitarian Evidence: $60 Million for the Whole Planet",[597],{"name":44,"to":45,"avatar":598},{"src":47},{"label":600},"Humanitarian Data | Sector Analysis",{"type":51,"value":602,"toc":1244},[603,606,609,613,616,748,755,759,911,915,918,962,966,975,984,994,1000,1007,1013,1019,1023,1026,1052,1059,1063,1074,1078,1081,1084,1090,1096,1103,1107,1110,1116,1123,1126,1132,1139,1144,1148,1155,1165,1174,1188,1198,1207,1210,1214,1220,1226,1232,1238],[58,604,605],{},"With the help of Claude, I pulled every edition of thirteen humanitarian flagship publications from 2022 to 2026 into one folder — sixty PDFs. Then I went looking for what they cost to produce, which turned out to be the more interesting half of the exercise.",[58,607,608],{},"This is a guide to the instruments we use to describe the sector every year, and which shape decisions organisations take: which one answers which question, how to quote them without getting caught out, where the holes are, and what the whole apparatus costs relative to what industry spends knowing about far smaller things.",[54,610,612],{"id":611},"the-publishing-calendar","The publishing calendar",[58,614,615],{},"Most of these reports are annual and they land in a predictable order. As a consequence, between January and April you are working with figures that are already a year old and an appeal written the previous November. And if you are drafting anything in May or June, wait: displacement, food security and funding all refresh within about six weeks of each other. Not a big deal, but something to have in mind.",[617,618,619,635],"table",{},[620,621,622],"thead",{},[623,624,625,629,632],"tr",{},[626,627,628],"th",{},"When",[626,630,631],{},"What lands",[626,633,634],{},"Covers",[636,637,638,650,665,678,696,706,716,734],"tbody",{},[623,639,640,644,647],{},[641,642,643],"td",{},"December",[641,645,646],{},"OCHA Global Humanitarian Overview; IRC Emergency Watchlist",[641,648,649],{},"The year ahead",[623,651,652,655,662],{},[641,653,654],{},"March–April",[641,656,657,658,661],{},"CRED ",[373,659,660],{},"Disasters in Numbers"," (EM-DAT)",[641,663,664],{},"Previous calendar year",[623,666,667,670,676],{},[641,668,669],{},"April–May",[641,671,672,673],{},"Global Report on Food Crises; IDMC ",[373,674,675],{},"Global Report on Internal Displacement",[641,677,664],{},[623,679,680,683,694],{},[641,681,682],{},"June",[641,684,685,686,689,690,693],{},"UNHCR ",[373,687,688],{},"Global Trends","; Global Humanitarian Assistance report; NRC ",[373,691,692],{},"most neglected crises","; GHO mid-year review",[641,695,664],{},[623,697,698,701,704],{},[641,699,700],{},"August",[641,702,703],{},"Aid Worker Security Report",[641,705,664],{},[623,707,708,711,714],{},[641,709,710],{},"October",[641,712,713],{},"Grand Bargain self-reporting highlights",[641,715,664],{},[623,717,718,721,731],{},[641,719,720],{},"Irregular",[641,722,723,724,727,728],{},"IFRC ",[373,725,726],{},"World Disasters Report","; UNDRR ",[373,729,730],{},"Global Assessment Report",[641,732,733],{},"Thematic",[623,735,736,739,745],{},[641,737,738],{},"Every four years",[641,740,741,742],{},"ALNAP ",[373,743,744],{},"State of the Humanitarian System",[641,746,747],{},"The previous four years",[58,749,750],{},[82,751],{"alt":752,"className":753,"src":754},"Grid of the thirteen publications by edition year, 2022 to 2026, showing full editions, partial or substitute editions, and years with nothing published",[86,87,76],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fflagship-coverage-2022-2026.png",[54,756,758],{"id":757},"which-report-answers-which-question","Which report answers which question",[617,760,761,774],{},[620,762,763],{},[623,764,765,768,771],{},[626,766,767],{},"Theme",[626,769,770],{},"Source \u002F publication",[626,772,773],{},"Notes and timing",[636,775,776,789,799,809,822,832,842,855,865,878,888,898],{},[623,777,778,783,786],{},[641,779,780],{},[65,781,782],{},"Needs, risk and crisis prioritisation",[641,784,785],{},"Global Humanitarian Overview (GHO)",[641,787,788],{},"December release with June mid-year review; an appeal document rather than a pure needs assessment",[623,790,791,793,796],{},[641,792],{},[641,794,795],{},"IRC Emergency Watchlist",[641,797,798],{},"Focuses on 20 countries, with the top 10 ranked",[623,800,801,803,806],{},[641,802],{},[641,804,805],{},"NRC Most Neglected Displacement Crises",[641,807,808],{},"Weighs funding, media coverage and political neglect; now incorporates scale",[623,810,811,816,819],{},[641,812,813],{},[65,814,815],{},"Sectoral crises and population movement",[641,817,818],{},"Global Report on Food Crises (GRFC)",[641,820,821],{},"Built systematically on IPC\u002FCH analyses",[623,823,824,826,829],{},[641,825],{},[641,827,828],{},"UNHCR Global Trends",[641,830,831],{},"Tracks end-of-year population stocks as well as annual movement flows",[623,833,834,836,839],{},[641,835],{},[641,837,838],{},"IDMC Global Report on Internal Displacement (GRID)",[641,840,841],{},"Distinguishes displaced people (stocks) from displacement movements (flows)",[623,843,844,849,852],{},[641,845,846],{},[65,847,848],{},"Disaster impact, risk reduction and economics",[641,850,851],{},"CRED Disasters in Numbers \u002F EM-DAT",[641,853,854],{},"Covers natural hazards only",[623,856,857,859,862],{},[641,858],{},[641,860,861],{},"UNDRR Global Assessment Report (GAR)",[641,863,864],{},"Released roughly biennially, alongside thematic special reports",[623,866,867,872,875],{},[641,868,869],{},[65,870,871],{},"Financing, operational performance and reform",[641,873,874],{},"Global Humanitarian Assistance (GHA) report",[641,876,877],{},"The only consolidated financing picture; produced by ALNAP since 2025",[623,879,880,882,885],{},[641,881],{},[641,883,884],{},"ALNAP State of the Humanitarian System (SOHS)",[641,886,887],{},"Comprehensive sector performance review, every four years",[623,889,890,892,895],{},[641,891],{},[641,893,894],{},"Grand Bargain reporting",[641,896,897],{},"Tracks system commitments; the independent annual review ended after 2023",[623,899,900,905,908],{},[641,901,902],{},[65,903,904],{},"Security and duty of care",[641,906,907],{},"Aid Worker Security Report (AWSR) \u002F AWSD",[641,909,910],{},"Annual report backed by an open, year-round queryable database",[167,912,914],{"id":913},"other-annual-publications-worth-tracking","Other annual publications worth tracking",[58,916,917],{},"These are not in the library, but they answer questions the thirteen do not.",[490,919,920,926,932,938,944,950,956],{},[493,921,922,925],{},[65,923,924],{},"UN Secretary-General — Protection of Civilians"," (annual, May). The UN's minimum count of civilian deaths in conflict.",[493,927,928,931],{},[65,929,930],{},"UN Secretary-General — Children and Armed Conflict"," (annual, June). Verified grave violations and the \"list of shame\" annexes.",[493,933,934,937],{},[65,935,936],{},"Save the Children — Stop the War on Children"," (annual). With PRIO, the standard source on children living in conflict zones.",[493,939,940,943],{},[65,941,942],{},"ACAPS — Humanitarian Access Overview"," (twice a year). The only systematic cross-country ranking of access constraints.",[493,945,946,949],{},[65,947,948],{},"INFORM Risk"," (EC JRC \u002F IASC; annual plus mid-year). The standard crisis-risk index used for prioritisation.",[493,951,952,955],{},[65,953,954],{},"CERF Annual Results Report"," (annual). Pooled-fund allocations, including the underfunded-emergency rounds.",[493,957,958,961],{},[65,959,960],{},"FAO et al. — State of Food Security and Nutrition in the World"," (SOFI, annual). Chronic hunger; complements the GRFC's acute focus.",[54,963,965],{"id":964},"six-rules-for-quoting-these-numbers","Six rules for quoting these numbers",[58,967,968,971,972],{},[65,969,970],{},"The edition year is not the data year."," The GHO 2026 was published in December 2025 and plans for 2026. The GRFC 2026 came out in April 2026 and describes 2025. NRC and CRED name editions after the data year, so NRC's 2025 list appeared in June 2026. ",[65,973,974],{},"Your citation has to distinguish publication from reference period.",[58,976,977,980,981],{},[65,978,979],{},"Figures are revised between editions."," International humanitarian assistance for 2022 appears as US$46.9bn in the GHA 2023, US$43.9bn in Development Initiatives' 2024 report, US$47.5bn in the SOHS 2026 executive summary and US$46.1bn in the same report's performance table. Deflators, revised FTS data, different inclusion rules. ",[65,982,983],{},"Quote a figure with its edition.",[58,985,986,989,990,993],{},[65,987,988],{},"Never mix editions inside one series."," Do not build a five-year trend by taking each year's figure from the edition published that year. Aid worker deaths are the clean example: 2023 was printed as 280 and is now 297, 2024 as 383 and now 387. A trend built from printed figures shows a smaller 2023→2024 jump than a trend built from the current database. ",[65,991,992],{},"Take every point in a series from the same source vintage"," — ideally the latest edition, which restates all prior years on one basis.",[58,995,996,999],{},[65,997,998],{},"Coverage changes move the trend line."," The GRFC's country set expanded from 48 to 73 for the 2023 edition, inflating the year-on-year rise. In 2026 the reverse: eighteen countries had no usable data, the lowest coverage in a decade, so part of the fall from 295 to 266 million people in acute food insecurity is missing data rather than improvement. IDMC says the same of its own 2025 figures — reduced data availability in fifteen per cent of monitored countries, triple the previous year.",[300,1001,1004],{"icon":1002,"title":1003},"i-ph-warning-diamond-duotone","Declines in a year of cuts should raise an alarm",[58,1005,1006],{},"When the measuring apparatus is losing funding at the same time as the thing it measures, a falling number is ambiguous by default. Check the coverage note before you write the sentence — this is more relevant now than it has been at any point in the period.",[58,1008,1009,1012],{},[65,1010,1011],{},"Methodology changes break series."," The Aid Worker Security Database began systematically recording detentions in 2025 and backfilled earlier years, so 2025's record total is partly definitional even though deaths fell. NRC added scale of displacement as a fourth criterion in 2026, which is largely why Sudan entered at number one.",[58,1014,1015,1018],{},[65,1016,1017],{},"Know which denominator you are using."," People in need, targeted, prioritised and reached are four different numbers. For 2026: 239 million in need, 135 million targeted, 87 million in the hyper-prioritised core. SOHS 2026 records the compression — the share of people in need who were targeted fell from 67% in 2023 to 60% in 2025, with 38% hyper-prioritised. Using \"people in need\" where you mean \"people we plan to reach\" overstates coverage by a factor of nearly three.",[54,1020,1022],{"id":1021},"where-the-holes-are","Where the holes are",[58,1024,1025],{},"Four of the thirteen series have gaps in this period, of four different kinds.",[490,1027,1028,1034,1040,1046],{},[493,1029,1030,1033],{},[65,1031,1032],{},"By design."," The SOHS appears every four years — ALNAP says the interval lets it capture major shifts while tracking long-term trends — and each edition covers the four years since the last, although this periodicity has changed since its inception. UNDRR alternates full GAR editions with special reports.",[493,1035,1036,1039],{},[65,1037,1038],{},"By governance decision."," The Grand Bargain's independent review, written by ODI\u002FHPG, ended after 2023 when signatories agreed to simplify reporting under the Grand Bargain 3.0 framework. What replaced it is a Secretariat summary of self-reported data that the Secretariat says it cannot verify. The Grand Bargain's mandate expires in 2026.",[493,1041,1042,1045],{},[65,1043,1044],{},"Unexplained."," IFRC published no World Disasters Report between the 2022 edition (which came out in January 2023) and March 2026.",[493,1047,1048,1051],{},[65,1049,1050],{},"By attrition."," The GHO 2026 exists only as an online e-report plus summary PDFs, in a year when OCHA cut about a fifth of its staff against a US$58 million shortfall. OCHA has not linked the two. But if you are building an archive, the full-report PDF you are used to filing no longer exists.",[58,1053,1054],{},[82,1055],{"alt":1056,"className":1057,"src":1058},"Grid of the thirteen publications by edition year showing which funders each edition acknowledges, with USAID-funded editions highlighted",[86,87,76],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fflagship-funders-2022-2026.png",[54,1060,1062],{"id":1061},"what-usaid-was-paying-for","What USAID was paying for",[58,1064,1065,1066,1069,1070,1073],{},"USAID co-funded the Global Report on Food Crises in 2022 and 2023, and was a named funding partner of IDMC's displacement report from 2022 to 2024. It appears in neither now. It paid for roughly ninety per cent of EM-DAT's €300,000 budget — ",[65,1067,1068],{},"the four-researcher operation behind almost every disaster mortality figure in our proposals","; CRED was notified in February 2025 and still had no replacement in writing fifteen months later. Humanitarian Outcomes states that the 2025 Aid Worker Security Report was almost not produced after its database lost US government funding. ",[65,1071,1072],{},"FEWS NET, four decades of famine early warning, went dark for six months."," Development Initiatives, which produced the Global Humanitarian Assistance report for twenty-five years, closed — ALNAP's own State of the Humanitarian System calls it a casualty of the cuts. And ALNAP, which inherited that report, had USAID as its single largest funder at twenty-eight per cent of income.",[54,1075,1077],{"id":1076},"what-the-whole-thing-costs","What the whole thing costs",[58,1079,1080],{},"Nobody appears to publish what the humanitarian sector spends on knowing things. Funding is coded by donor, recipient, crisis and channel — never by function — so there is no \"assessment, information management and analysis\" line anywhere in FTS, OECD DAC or the GHA report's own taxonomy.",[58,1082,1083],{},"So I reconstructed it from published accounts, grant records and procurement data, in two layers.",[58,1085,1086,1089],{},[65,1087,1088],{},"Producing the thirteen reports costs roughly US$8–17 million a year",", midpoint about US$12 million. Some of that is anchored in the publishers' own accounts: IDMC spent US$7.0 million in 2023 with 64 staff; ALNAP £2.0 million in 2024–25 with 17 staff, producing both the SOHS and, since 2025, the GHA report; Humanitarian Outcomes received CAD 400,000 from Canada for sixteen months of core research including the Aid Worker Security Database; CRED runs EM-DAT on about €300,000 a year with four researchers. The rest — the GHO, the GAR — is estimated, because OCHA and UNDRR have never published what their flagships cost.",[58,1091,1092,1095],{},[65,1093,1094],{},"The data systems underneath add about US$52 million a year",", and here the figures are mostly public, because they are grant-funded programmes rather than in-house activities: FEWS NET at roughly US$30 million a year (US$213 million obligated between 2019 and 2026), the IPC's global support programme at about US$8 million, IDMC's monitoring operation, the World Bank–UNHCR Joint Data Center at US$4.7 million, and OCHA's Centre for Humanitarian Data at around US$1.8 million.",[300,1097,1100],{"icon":1098,"title":1099},"i-ph-coins-duotone","On the order of US$60–70 million a year",[58,1101,1102],{},"Against the 2025 appeal of US$47 billion that is about 0.13%; against the US$33.3 billion actually spent, about 0.19%. And it is a floor — it excludes needs assessments, the JIAF, cluster information staff, REACH's multi-sector assessments, UNHCR's statistics function and agency monitoring, none of which publish a figure.",[54,1104,1106],{"id":1105},"what-the-world-spends-knowing-about-other-things","What the world spends knowing about other things",[58,1108,1109],{},"What else in the world costs about US$60 million a year to know about?",[58,1111,1112,1115],{},[65,1113,1114],{},"Barb Audiences, the company that measures what British households watch on television, ran on £40.7 million in 2024 with fourteen employees and a seven-thousand-home panel."," Knowing what one country watches on TV costs roughly four-fifths of what the world spends on its entire shared humanitarian evidence base.",[58,1117,1118],{},[82,1119],{"alt":1120,"className":1121,"src":1122},"Horizontal bar chart comparing annual spend on producing analysis about one domain, from global coffee statistics at three million dollars to US media audience measurement at two hundred and fifty-four million, with the humanitarian evidence base at sixty-two million",[86,87,76],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fcost-of-knowing-by-subject.png",[58,1124,1125],{},"Just above the line: WGSN, which tells clothing brands what colours will be fashionable in eighteen months, turned over £91.3 million at a 45% EBITDA margin. Mintel and Euromonitor, two publishers of consumer market reports, took about US$200 and US$220 million. Comscore's syndicated audience business, counting who watched which advertisement, took US$254 million. On the public side, the United States spends US$135 million a year on energy statistics through the EIA and US$188 million on crop and livestock statistics through USDA NASS — each, for one country and one commodity class, more than the world spends on humanitarian analysis for everywhere.",[58,1127,1128,1129],{},"The UK Office for National Statistics cost £374.8 million in 2024–25. Sportradar, selling sports data to bookmakers, earned €1.29 billion. L'Oréal spent €1.38 billion on research and innovation with four thousand scientists — ",[65,1130,1131],{},"about twenty-two times the humanitarian evidence base, on cosmetics.",[58,1133,1134,1135,1138],{},"In the other direction, to avoid dramatising: the International Coffee Organization maintains global coffee statistics on £2.7 million a year with fifteen posts, and the International Cocoa Organization runs global cocoa statistics on €3.3 million, partly funding itself by selling the data. Both are an order of magnitude ",[373,1136,1137],{},"below"," the humanitarian evidence base.",[58,1140,1141],{},[65,1142,1143],{},"A system this cheap is not robust, and it breaks when one funder leaves.",[54,1145,1147],{"id":1146},"the-five-year-review","The five-year review",[58,1149,1150,1151,1154],{},"Below is what each year's editions actually said. Years here are ",[65,1152,1153],{},"edition years"," (the year of publication). Each set describes the previous calendar year, except the GHO and the IRC Watchlist, which look forward.",[58,1156,1157,1160,1161,1164],{},[65,1158,1159],{},"2022 editions — reporting on 2021, published as Ukraine began."," The GHO asked for US$41 billion for 183 million of 274 million people in need, up from 235 million a year earlier. UNHCR counted 89.3 million forcibly displaced at end-2021 and IDMC a then-record 59.1 million IDPs, with Ethiopia's 5.1 million conflict displacements the highest ever recorded for a single country in one year. The GRFC put 193 million people in IPC Phase 3 or worse — nearly 40 million more than 2020 — and 570,000 in Catastrophe across Ethiopia, South Sudan, Yemen and southern Madagascar. ",[65,1162,1163],{},"Funding was flat at US$31.3 billion, and direct funding to local actors fell to 1.2%, the lowest in the series."," The SOHS 2022 reported that the system reached about 46% of the people it had itself identified as in need.",[58,1166,1167,1170,1171],{},[65,1168,1169],{},"2023 editions — reporting on 2022: Ukraine, the Horn of Africa drought, the largest displacement increase ever recorded."," The GHO launched at a record US$51.5 billion for 230 million of 339 million in need — 65 million more people than the year before. Forced displacement rose 21% to 108.4 million; the IDP stock rose 20% to 71.1 million on a record 60.9 million internal displacements, of which Ukraine accounted for roughly 16.9 million movements and Pakistan's floods 8.2 million. Food insecurity rose for a fourth consecutive year to 258 million, with 214,100 of the 376,400 people in Catastrophe in Somalia alone. ",[65,1172,1173],{},"International humanitarian assistance hit an all-time high of US$46.9 billion while the appeals shortfall hit an all-time high of US$22.1 billion.",[58,1175,1176,1179,1180,1183,1184,1187],{},[65,1177,1178],{},"2024 editions — reporting on 2023: Sudan, Gaza, and the first deliberately leaner appeal."," The GHO asked for less than the previous year — US$46.4 billion for 180.5 million of 299.4 million in need — the first explicit narrowing of ambition. Gaza accounted for 576,600 of the 705,000 people in IPC Catastrophe, the highest total in GRFC history, and for 3.4 million displacements in the last quarter of 2023 alone. Sudan reached 9.1 million IDPs, the most ever recorded for one country. ",[65,1181,1182],{},"Humanitarian Outcomes counted 280 aid workers killed — the deadliest year on record at the time — 163 of them in Gaza in the war's first three months",", ",[65,1185,1186],{},"with 57% of all deaths caused by aerial bombardment, a method almost exclusively available to states."," Funding stalled at US$43.4 billion (−1.1%), appeal coverage fell to 45%, and Development Initiatives projected a further 11% fall.",[58,1189,1190,1193,1194,1197],{},[65,1191,1192],{},"2025 editions — reporting on 2024, published into the collapse."," Every indicator set a record in the wrong direction: 83.4 million IDPs, above 80 million for the first time; 1.9 million people in Catastrophe, the highest since GRFC reporting began; 383 aid workers killed; Sudan at 11.6 million IDPs; food insecurity up for a sixth consecutive year to 295.3 million. ALNAP's first GHA edition reported that assistance had fallen 10% in 2024 — \"the largest cut ever recorded\" — and modelled public donor funding down 34–45% from the 2023 peak. Then the January 2025 US freeze arrived mid-cycle. OCHA rewrote the appeal in June under the title ",[373,1195,1196],{},"The Cruel Math of Aid Cuts",": 114.4 million people identified for hyper-prioritised assistance out of 305 million in need, needing US$29.1 billion — of which 18.5% had been received by 10 June.",[58,1199,1200,1203,1204],{},[65,1201,1202],{},"2026 editions — reporting on 2025: a smaller system."," The GHO was built around hyper-prioritisation from the start: 239 million in need, 135 million targeted, 87 million in the prioritised core, US$33 billion requested, about 30% below the 2025 ask. Assistance fell 20% to US$33.3 billion, the US share of government funding dropped to about 23%, and the UAE and Saudi Arabia entered the top five donors. Two famines were confirmed in the same year, in Gaza and Sudan. For the first time in a decade both global forced displacement and the IDP stock declined — but largely through returns under duress and through data that no longer exists: the GRFC 2026 has its lowest coverage in a decade, with eighteen countries lacking usable analysis, so part of the fall from 295 to 266 million is arithmetic rather than improvement. The SOHS 2026 supplies the summary judgement: the share of people in need who were even targeted fell from 67% in 2023 to 60% in 2025, and ",[65,1205,1206],{},"the system entered 2026 smaller, more fragile, and facing open questions about its role.",[58,1208,1209],{},"The 274 million people in need in the 2022 edition and the 239 million in the 2026 edition are not the same measurement. Needs did not fall by 35 million; the definition of who counts tightened under funding pressure, the GHO moved from planning for everyone in need to planning for a prioritised core, and assessment coverage shrank. The series that runs most cleanly is the funding one, because it is counted rather than estimated — and taken from a single edition, the SOHS 2026, it reads as a peak of US$46.1 billion in 2022, a 13% fall in 2024 and a further 20% in 2025, to US$33.3 billion. Note that this is not the same 2024 decline as the 10% reported by the GHA 2025 above: different edition, revised data, same direction.",[54,1211,1213],{"id":1212},"what-i-would-actually-do-with-this","What I would actually do with this",[58,1215,1216,1219],{},[65,1217,1218],{},"Keep local copies."," Publisher URLs rot faster than you expect: the FSIN links for older GRFC editions are dead, and the only reliable copy of the 2022 report I found was in a CGIAR repository. Download the PDF on publication day, one folder per series.",[58,1221,1222,1225],{},[65,1223,1224],{},"Cite the edition, not the report."," \"GRFC 2025, p. 12\" ages well. \"The GRFC says 295 million\" will be changed in the next revision.",[58,1227,1228,1231],{},[65,1229,1230],{},"Go to the annexes and the data."," The GHO master dataset is on HDX, EM-DAT and the AWSD are queryable, IDMC publishes country tables, and the GHA figures come from FTS and OECD DAC. For anything you will reuse, take the number from the data and use the report for the framing.",[58,1233,1234,1237],{},[65,1235,1236],{},"Write down which denominator you used."," In the same year, the gap between people in need and people prioritised is 152 million, so it does matter to say what you are counting.",[560,1239],{"description":1240,"icon":1241,"link-label":1242,"link-url":565,"title":1243},"Thirteen series, sixty editions, 2022 to 2026, with a year-by-year summary, the funding and coverage charts and the cost workings. Assembled for internal use — happy to share it as a reference for your team.","i-ph-books-duotone","Get in touch","Want the library?",{"title":248,"searchDepth":249,"depth":249,"links":1245},[1246,1247,1250,1251,1252,1253,1254,1255,1256],{"id":611,"depth":249,"text":612},{"id":757,"depth":249,"text":758,"children":1248},[1249],{"id":913,"depth":255,"text":914},{"id":964,"depth":249,"text":965},{"id":1021,"depth":249,"text":1022},{"id":1061,"depth":249,"text":1062},{"id":1076,"depth":249,"text":1077},{"id":1105,"depth":249,"text":1106},{"id":1146,"depth":249,"text":1147},{"id":1212,"depth":249,"text":1213},"2026-09-14T00:00:00.000Z","A practitioner's guide to the humanitarian sector's flagship publications — which one answers which question, how to quote them without getting caught out, and what the whole evidence base costs compared with what the world spends knowing about television ratings.",{"src":1260,"alt":1261},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00053_.png","Hand-drawn illustration of a tall stack of annual reports with two groups of people standing on either side of it",{},"\u002Farticles\u002Fhumanitarian-flagship-reports",{"title":595,"description":1258},"3.articles\u002F11.humanitarian-flagship-reports","Y0R-q264ixMDnz2yWTDocG3JxnFLwyuzmfO-9O1eHag",{"id":1268,"title":1269,"authors":1270,"badge":1273,"body":1275,"date":1396,"description":1397,"extension":266,"external_links":1398,"image":1403,"meta":1406,"navigation":269,"path":1407,"seo":1408,"stem":1409,"__hash__":1410},"posts\u002F3.articles\u002F2.chat-agent.md","The Humanitarian Assistant is here again, completely revamped",[1271],{"name":44,"to":45,"avatar":1272},{"src":47},{"label":1274},"AI",{"type":51,"value":1276,"toc":1390},[1277,1280,1287,1291,1297,1300,1304,1311,1322,1326,1329,1369,1372,1376,1387],[58,1278,1279],{},"In the humanitarian sector, new field managers are often thrust into roles demanding an unrealistic breadth of knowledge. They are expected to instantly master complex donor regulations, reporting procedures, humanitarian principles, and internal policies, leaving them in delicate positions without immediate support. This challenge highlights a critical gap for the dedicated individuals on the front lines.",[58,1281,1282,1283,1286],{},"To address this, I built the ",[65,1284,1285],{},"Humanitarian Assistant",", a dynamic AI tool designed to act as a virtual colleague. It is more than just a chatbot; it is a real-time support system built to provide the contextually relevant assistance that humanitarian workers need to succeed—grounded in trusted public sources and honest about where its answers come from.",[54,1288,1290],{"id":1289},"designing-for-immediate-access","Designing for Immediate Access",[58,1292,1293,1294,1296],{},"We believe that in a crisis, barriers to information should not exist. That is why the Humanitarian Assistant follows a strict \"Open Access\" philosophy. Any visitor can land on the page and immediately start asking questions—no login required. However, for those who need to manage long-term projects or revisit complex reasoning chains, the tool integrates seamlessly with ",[65,1295,67],{},". By signing in, users can save their conversation history, effectively building a digital memory bank of their work.",[58,1298,1299],{},"Rather than asking users to pick the right \"mode\" for every question, the assistant works like a versatile team member: it reads your question and automatically reaches for the right source—a knowledge library, a live data feed, or a web search—and combines them into a single, coherent answer.",[54,1301,1303],{"id":1302},"grounded-in-public-humanitarian-knowledge","Grounded in Public Humanitarian Knowledge",[58,1305,1306,1307,1310],{},"In our sector, data privacy and the \"Do No Harm\" principle are non-negotiable. We made a conscious architectural decision to build the knowledge base exclusively from ",[65,1308,1309],{},"publicly available, verified humanitarian guidance","—ALNAP lessons and evaluations, IASC guidance, the cash and voucher assistance (CVA) toolkit, ECHO eligibility rules, and child-protection and PSEA standards, among others. We do not scrape private internal datasets or sensitive operational files.",[58,1312,1313,1314,1317,1318,1321],{},"These documents—hundreds of reports, distilled into thousands of searchable passages—are embedded and stored in a ",[65,1315,1316],{},"self-hosted Qdrant vector database"," that we run and control end to end. When you ask a question, the assistant retrieves the most relevant passages and ",[65,1319,1320],{},"cites the source documents"," it drew on, so you can verify the answer and read further. By owning the retrieval layer rather than relying on opaque third-party services, we keep every answer transparent and traceable to public-sector knowledge.",[54,1323,1325],{"id":1324},"live-data-not-just-documents","Live Data, Not Just Documents",[58,1327,1328],{},"Static guidance is only half the picture, so the assistant also queries live humanitarian data sources on demand:",[490,1330,1331,1349,1363],{},[493,1332,1333,1336,1337,1340,1341,1344,1345,1348],{},[65,1334,1335],{},"ACAPS"," — displacement and people-in-need figures for Ukraine (down to the oblast level), plus global crisis analytics: the ",[65,1338,1339],{},"INFORM Severity Index",", the ",[65,1342,1343],{},"Risk List"," of emerging threats, and ",[65,1346,1347],{},"Humanitarian Access"," constraints for any country.",[493,1350,1351,1354,1355,1358,1359,1362],{},[65,1352,1353],{},"FEWS NET"," — acute food insecurity ",[65,1356,1357],{},"IPC phase classifications"," (from Minimal to Famine) and ",[65,1360,1361],{},"staple food market prices"," with year-on-year trends, anywhere in the world.",[493,1364,1365,1368],{},[65,1366,1367],{},"Web search"," for current events when the curated sources don't cover a question.",[58,1370,1371],{},"The agent decides which tool fits each question: a query about famine risk in the Horn of Africa pulls live IPC data and cites FEWS NET, while a question about cash programming draws on the knowledge base, and a request for displacement figures in a Ukrainian oblast hits ACAPS directly.",[54,1373,1375],{"id":1374},"orchestrated-intelligence-under-the-hood","Orchestrated Intelligence Under the Hood",[58,1377,1378,1379,1382,1383,1386],{},"While the interface is simple, the engineering behind it is deliberate. The assistant is built on ",[65,1380,1381],{},"Flowise",", orchestrating a ",[65,1384,1385],{},"Google Gemini"," model together with the retrieval and live-data tools above. We recently re-platformed the system off the OpenAI Assistants API onto Gemini and our own Qdrant vector store—a move that gives us independence over the model, full ownership of the retrieval layer, and a lower running cost. Hardened guardrails keep the assistant strictly on humanitarian topics and resistant to prompt-injection attempts.",[58,1388,1389],{},"The result is a secure, transparent, and cost-effective colleague—free for everyone to use, from international coordinators to local NGO partners.",{"title":248,"searchDepth":249,"depth":249,"links":1391},[1392,1393,1394,1395],{"id":1289,"depth":249,"text":1290},{"id":1302,"depth":249,"text":1303},{"id":1324,"depth":249,"text":1325},{"id":1374,"depth":249,"text":1375},"2025-05-01T00:00:00.000Z","A chat agent designed to support humanitarian workers with real-time, context-aware assistance using publicly available data.",[1399],{"label":1400,"url":1401,"icon":1402},"Humanitarian Agent Demo","\u002Fdemos\u002Fhumanitarian-agent","i-ph-robot",{"src":1404,"alt":1405},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00352_.png","Illustration of a friendly robot assisting a human worker",{},"\u002Farticles\u002Fchat-agent",{"title":1269,"description":1397},"3.articles\u002F2.chat-agent","NT4xE2y3yJ1nW8zhN0bkt-0WoSWX8U2TSjzpLdmXBAQ",{"id":1412,"title":1413,"authors":1414,"badge":1417,"body":1419,"date":1741,"description":1742,"extension":266,"external_links":1743,"image":1752,"meta":1754,"navigation":269,"path":1755,"seo":1756,"stem":1757,"__hash__":1758},"posts\u002F3.articles\u002F3.jobs-relief.md","Data Insights into the 2025 Humanitarian Sector Contraction",[1415],{"name":44,"to":45,"avatar":1416},{"src":47},{"label":1418},"Business Intelligence | AI",{"type":51,"value":1420,"toc":1724},[1421,1425,1435,1442,1446,1453,1464,1468,1471,1488,1498,1513,1523,1527,1530,1557,1572,1578,1581,1585,1592,1598,1601,1608,1612,1615,1628,1631,1634,1641,1645,1648,1657,1661,1665,1668,1672,1675,1678,1681,1685,1688,1691,1703,1707,1714,1721],[54,1422,1424],{"id":1423},"introduction","Introduction",[58,1426,1427,1428,441,1431,1434],{},"The modern humanitarian professional is a figure born from the crucible of past failures. The volunteer-dependent responses to crises in ",[65,1429,1430],{},"Biafra (1970)",[65,1432,1433],{},"Rwanda (1994)"," starkly revealed that good intentions alone were insufficient. These events catalyzed a shift, underscoring the necessity for a corps of professionals equipped with technical skills and specialized knowledge, not just a vocational calling for humanity.",[58,1436,1437,1438,1441],{},"This evolution spurred a wave of ",[65,1439,1440],{},"institutionalization and standardization"," from the 1990s onward, establishing the benchmarks for a new kind of aid worker. The result is the contemporary humanitarian: a managerial and technical expert, often with a strong administrative background. This professional identity is defined by a complex, and often conflicting, blend of technocratic capabilities and a deeply ingrained set of humanitarian ideals.",[167,1443,1445],{"id":1444},"the-current-crisis-the-great-contraction-of-2025","The Current Crisis: The Great Contraction of 2025",[58,1447,1448,1449,1452],{},"The year 2025 is being defined by a severe ",[65,1450,1451],{},"contraction of the humanitarian sector",". This downturn was triggered by the United States' decision on January 24th to freeze humanitarian funding with immediate effect. This move was quickly emulated by other key donor countries, including the United Kingdom, Germany, France, the Netherlands, Switzerland, and Sweden.",[58,1454,1455,1456,1459,1460,1463],{},"This crisis, however, did not emerge from a vacuum. It is an acceleration of a trend that began in 2024. For over a decade, the growth in humanitarian funding, ",[65,1457,1458],{},"while always failing to meet the ever-expanding global needs, was at least steady",". Last year marked a turning point, with total funds dropping by ",[65,1461,1462],{},"11%","—the first such decrease in more than ten years.",[54,1465,1467],{"id":1466},"the-human-component-workforce-dynamics-and-wellbeing","The Human Component: Workforce Dynamics and Wellbeing",[58,1469,1470],{},"To understand the full impact of this contraction, one must look beyond the budget lines to the demographics and operational realities of the workforce itself. The sector is characterized by distinct patterns of entry, retention, and attrition that are likely to be exacerbated by the current crisis.",[58,1472,1473,1476,1477,1480,1481,1484,1485,320],{},[65,1474,1475],{},"Age and Career Lifecycle","\nThe humanitarian workforce typically exhibits a bimodal age distribution. There is a significant peak of entry-level professionals in the ",[65,1478,1479],{},"25–35 age range",", driven by early-career enthusiasm and volunteerism. A second, smaller peak exists for senior experts aged ",[65,1482,1483],{},"55–65",", often returning to the field after other careers. Between these two groups lies a \"missing middle.\" Data suggests a high attrition rate as workers reach their mid-30s, often due to the incompatibility of field life with raising a family. Consequently, the \"active life\" of a field worker is surprisingly short; for many, it is less than a decade. Surveys indicate that over 45% of humanitarian professionals expect to remain in the sector for ",[65,1486,1487],{},"10 years or less",[58,1489,1490,1493,1494,1497],{},[65,1491,1492],{},"The Physical and Emotional Weight","\nThe toll of this work is measurable and severe. The physical weight of the profession has increased as the security environment deteriorates; 2024 was recorded as the deadliest year in history for aid workers, with ",[65,1495,1496],{},"281 fatalities"," globally, driven largely by conflicts in Gaza and Sudan.",[58,1499,1500,1501,1504,1505,1508,1509,1512],{},"The emotional weight is equally heavy. The prevalence of ",[65,1502,1503],{},"Post-Traumatic Stress Disorder (PTSD)"," among aid workers is estimated at up to ",[65,1506,1507],{},"30%"," upon return from assignment. Furthermore, burnout rates range significantly, with some studies showing between ",[65,1510,1511],{},"8.5% and 32%"," of workers suffering from emotional exhaustion. Alarmingly, health data suggests that 28% of workers report a deterioration in their physical health following a mission, often persisting months after their return.",[58,1514,1515],{},[290,1516,1519],{"href":1517,"rel":1518},"https:\u002F\u002Fbaena.ai\u002Fdemos\u002Freliefjobs-dashboard",[294],[82,1520],{"alt":1521,"src":1522},"A live dashboard tracking the state of humanitarian jobs","https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fdashboard-link.png",[54,1524,1526],{"id":1525},"how-to-monitor-the-situation-a-dashboard-approach","How to Monitor the Situation: A Dashboard Approach",[58,1528,1529],{},"To track the impact of these funding cuts on the workforce, I have developed a dashboard that provides a daily, high-level overview of the humanitarian job market.",[58,1531,1532,1533,1538,1539,441,1542,1545,1546,1551,1552,320],{},"The dashboard automatically collects and processes job posting data using the API from ",[290,1534,1537],{"href":1535,"rel":1536},"https:\u002F\u002Freliefweb.int\u002Fjobs",[294],"ReliefWeb",", one of the sector's primary information hubs. It then visualizes key trends, allowing for a quick assessment of how the market is changing. The underlying ",[65,1540,1541],{},"SQL queries",[65,1543,1544],{},"automation flows"," used to power this tool are open-source and available for review in this ",[290,1547,1550],{"href":1548,"rel":1549},"https:\u002F\u002Fgithub.com\u002FJesus-Baena\u002F005-dashboard-reliefweb-jobs",[294],"GitHub repository"," and this entry of my blog: ",[290,1553,1556],{"href":1554,"rel":1555},"https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FB.-HUMANITARIAN-SECTOR-EMPLOYMENT-DASHBOARD",[294],"Building a Humanitarian Jobs Dashboard",[300,1558,1561],{"icon":1559,"title":1560},"i-ph-warning-duotone","A Major Disclaimer",[58,1562,1563,1564,1567,1568,1571],{},"While invaluable for a ",[373,1565,1566],{},"\"sneak peek\""," into market dynamics, ",[65,1569,1570],{},"a dashboard has its limits",". It reveals trends and correlations but cannot provide the rigorous, in-depth analysis needed to draw definitive conclusions. The interpretations that follow are therefore informed hypotheses, not final judgments.",[58,1573,1574,1577],{},[65,1575,1576],{},"Update — August 2026."," The dashboard has been rebuilt. What you see embedded below is a Nuxt front end, not an embedded Metabase dashboard. The pipeline behind it is unchanged — ReliefWeb API → n8n → Postgres, with LLM extraction on the job descriptions — and the queries are the same ones. Only the presentation layer moved. The reasoning is at the end of this article.",[1579,1580],"hr",{},[54,1582,1584],{"id":1583},"interpretation-reading-the-data","Interpretation: Reading the Data",[58,1586,1587,1588,1591],{},"The impact of the 2025 funding crisis is clear and immediate. ",[65,1589,1590],{},"Monthly job postings have been in a sustained downturn"," since the February funding freeze.",[58,1593,1594],{},[82,1595],{"alt":1596,"src":1597},"Chart Downturn jobs trend","https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fjobs-trend.png",[58,1599,1600],{},"While a minor recovery in postings was observed in July, this more likely signals that the market is beginning to normalize at a new, much lower baseline rather than indicating a true recovery.",[58,1602,1603,1604,1607],{},"Notably, ",[65,1605,1606],{},"short-term consultancy roles appear largely impervious"," to this downward trend. This is a critical finding. It suggests a potential strategic shift in hiring, with organizations leaning on flexible, short-term contracts to navigate the budget crisis.",[167,1609,1611],{"id":1610},"the-acceleration-of-precarity","The Acceleration of Precarity",[58,1613,1614],{},"This pivot towards consultancy is not just a temporary fix; it's an acceleration of the sector's most problematic structural habits. The humanitarian field has long been criticized for practices that hurt aid effectiveness, such as a reliance on short-term contracts, which fuels labor discontinuity and a constant \"brain drain.\"",[58,1616,1617,1620,1621,1624,1625,320],{},[65,1618,1619],{},"The dominance of one-year, project-based funding cycles"," limits long-term planning, capacity building, and the overall quality of delivery. Standard assignment lengths for organizations like MSF or the ICRC typically range from ",[65,1622,1623],{},"6 to 12 months"," for first-time field staff, creating a constant churn of personnel. Research indicates that retention is a major challenge, with some studies showing that only ",[65,1626,1627],{},"40% of employees reenlist for a second mission",[58,1629,1630],{},"More often we see organizations rotating even senior management at the mission level, which often attracts short-term professionals with little commitment to the organization's mission or the communities they serve.",[58,1632,1633],{},"The new data suggests the 2025 crisis is not solving these issues but rather entrenching them. By favoring flexible (and precarious) gig-based roles over stable, full-time positions, organizations are trading long-term institutional stability for short-term financial survival. We are witnessing a transformative reduction in stable relief jobs, which could hollow out the professionalized core of the sector.",[58,1635,1636,1637,1640],{},"(",[373,1638,1639],{},"Note on Scope:"," While the USAID funding freeze was the catalyst for this crisis, this analysis will not focus on that specific event. There is already excellent, detailed reporting on the 'stop work' order and its immediate fallout from sources like ICVA and specialist platforms like usaidstopwork.com. This article's aim is to analyze the secondary, sector-wide workforce impact as seen through the job market data.)",[54,1642,1644],{"id":1643},"what-data-says-about-localization","What data says about Localization?",[58,1646,1647],{},"Contrary to some expectations, the data does not indicate a significant shift towards local hiring in response to the funding crisis. The proportion of job postings targeting local versus international candidates has remained relatively stable throughout the downturn.",[58,1649,1650,1651,1656],{},"The way this was calculated can be found here: ",[290,1652,1655],{"href":1653,"rel":1654},"https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FLLM-extraction-of-new-information-in-Job-Descriptions-about-localization",[294],"LLM Data Extraction from Job Descriptions",". I have used an LLM to read through the job descriptions and extract whether the position is open to local candidates only, international candidates only, or both.",[54,1658,1660],{"id":1659},"the-technology-behind-this-project","The technology behind this project",[167,1662,1664],{"id":1663},"metabase-an-open-source-alternative-to-power-bi","Metabase, an open-source alternative to Power BI",[58,1666,1667],{},"Metabase carried this project for its first year and deserves the credit. Pointing it at a Postgres instance takes minutes, the query builder is genuinely usable by people who don't write SQL, and self-hosting it costs nothing beyond the container it runs in. For getting from \"I have a table\" to \"I have a dashboard someone else can read,\" there is very little that beats it.",[167,1669,1671],{"id":1670},"why-the-front-end-moved-to-nuxt","Why the front end moved to Nuxt",[58,1673,1674],{},"Where it stops being the right tool is presentation. An embedded dashboard brings its own layout, typography, and chart defaults into the host page, and none of those are yours. On a public-facing article the seams show: the iframe doesn't inherit the site's type or palette, the desktop grid squeezes rather than reflows on a phone, and every visitor waits for a second application to boot inside the first one. Dense charts that read fine in an analyst's browser tab become unreadable on the device most people actually open the link on.",[58,1676,1677],{},"So the front end was rebuilt in Nuxt. Same pipeline, same SQL, same rolling seven-day logic — but served through an API layer and rendered as native components. The charts inherit the site's typography and colours, the layout is responsive by design rather than scaled down, and the page loads as one application instead of two. Metabase stays in the stack for exploration and ad-hoc querying, which is where it is genuinely best. The published view is now something I control end to end.",[58,1679,1680],{},"This is the general pattern, not a criticism of the tool: BI platforms are built for analysts working inside them, not for readers arriving from a link. If the audience is internal, embed and move on. If the dashboard is the public artefact, the presentation layer is worth owning.",[167,1682,1684],{"id":1683},"methodological-notes-the-rolling-week-problem","Methodological Notes: The \"Rolling Week\" Problem",[58,1686,1687],{},"Building and interpreting a live dashboard comes with its own set of technical challenges. Metabase excels at composing queries based on fixed calendar ranges (e.g., \"Last Week\" or \"This Month\"). However, this creates analytical noise.",[58,1689,1690],{},"For example, when viewing the dashboard at the beginning of a new week (a Monday or Tuesday), comparing the few jobs posted so far to the full seven days of the previous week results in a disproportionate and misleadingly alarming drop.",[58,1692,1693,1694,1697,1698],{},"The solution is to implement a ",[373,1695,1696],{},"\"rolling seven-day total\""," query. This provides a far more accurate and stable picture of the immediate trend, smoothing out the noise from arbitrary calendar cutoffs. This is one of several small but critical nuances required to make the data tell an honest story. The query example can be found here: ",[290,1699,1702],{"href":1700,"rel":1701},"https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FThe-rolling-week-problem",[294],"Rolling Week SQL Query",[54,1704,1706],{"id":1705},"conclusion-a-sector-at-a-crossroads","Conclusion: A Sector at a Crossroads",[58,1708,1709,1710,1713],{},"The data points to a sector not just shrinking, ",[65,1711,1712],{},"but fundamentally changing its employment structure",". The convergence of the 2024 funding drop and the 2025 funding freeze is forcing a painful, sector-wide recalibration.",[58,1715,1716,1717,1720],{},"The resilience of consultancy roles amidst a collapse in full-time positions suggests a structural shift towards a more transient, ",[65,1718,1719],{},"\"gig-based\" humanitarian workforce",". While this may address immediate budget shortfalls, it threatens to deepen the long-standing issues of precarity, knowledge drain, and inconsistent quality that the sector has fought for two decades to overcome.",[58,1722,1723],{},"The professionalized, technical expert—born from the failures of the 1990s—is now facing a new, existential storm: not just a lack of funds, but a hollowing out of the very institutional stability that allowed the profession to mature in the first place.",{"title":248,"searchDepth":249,"depth":249,"links":1725},[1726,1729,1730,1731,1734,1735,1740],{"id":1423,"depth":249,"text":1424,"children":1727},[1728],{"id":1444,"depth":255,"text":1445},{"id":1466,"depth":249,"text":1467},{"id":1525,"depth":249,"text":1526},{"id":1583,"depth":249,"text":1584,"children":1732},[1733],{"id":1610,"depth":255,"text":1611},{"id":1643,"depth":249,"text":1644},{"id":1659,"depth":249,"text":1660,"children":1736},[1737,1738,1739],{"id":1663,"depth":255,"text":1664},{"id":1670,"depth":255,"text":1671},{"id":1683,"depth":255,"text":1684},{"id":1705,"depth":249,"text":1706},"2025-06-01T00:00:00.000Z","A live dashboard to understand the state of humanitarian jobs in 2025",[1744,1748],{"label":1745,"url":1746,"icon":1747},"Jobs Analytics Dashboard (rebuilt)","\u002Fdemos\u002Freliefjobs-dashboard","i-ph-chart-bar",{"label":1749,"url":1750,"icon":1751},"GitHub Repository","https:\u002F\u002Fgithub.com\u002FJesus-Baena\u002F2025-reliefweb-jobs-dashboard","i-simple-icons-github",{"src":1753},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00250_.png",{},"\u002Farticles\u002Fjobs-relief",{"title":1413,"description":1742},"3.articles\u002F3.jobs-relief","cnaQTzBv6HnQYxXtreVPw-b1osaXUrH7R-I_IQ68JP8",{"id":1760,"title":1761,"authors":1762,"badge":1765,"body":1767,"date":2021,"description":2022,"extension":266,"external_links":2023,"image":2028,"meta":2031,"navigation":269,"path":2032,"seo":2033,"stem":2034,"__hash__":2035},"posts\u002F3.articles\u002F4.paralegal-advisor.md","An AI Legal Adviser for NGOs in Ukraine",[1763],{"name":44,"to":45,"avatar":1764},{"src":47},{"label":1766},"AI | LegalTech",{"type":51,"value":1768,"toc":2013},[1769,1778,1785,1793,1800,1804,1811,1815,1826,1846,1850,1857,1863,1869,1879,1885,1891,1898,1902,1913,1917,1920,1974,1978,1985,1996,1999,2001,2008],[58,1770,1771,1772,1777],{},"When I returned to Ukraine in my role as ",[290,1773,1776],{"href":1774,"rel":1775},"https:\u002F\u002Fbaena.blog\u002F00.-STOCK\u002FFeb-2025---Field-Coordinator-with-Doctors-of-the-World-in-UKRAINE",[294],"Field Coordinator with Doctors of the World"," in 2025, I took the opportunity to deepen my understanding of the local legislative landscape. I gathered and curated legislation—sometimes translated from Ukrainian to English—and built an assistant I could actually hold a conversation with about legal questions on the ground.",[58,1779,1780,1781,1784],{},"I'm now sharing it publicly as a ",[65,1782,1783],{},"free legal advisor for humanitarian NGOs",", designed to help both local and international organizations navigate Ukraine's legal framework. Country directors, operations managers, and technical coordinators will get the most out of it, and the knowledge base is especially rich in the health sector.",[58,1786,1787,1788],{},"Try it here: ",[290,1789,1792],{"href":1790,"rel":1791},"https:\u002F\u002Fbaena.ai\u002Fdemos\u002Fparalegal-advisor",[294],"baena.ai\u002Fdemos\u002Fparalegal-advisor",[300,1794,1797],{"icon":1795,"title":1796},"i-ph-info-duotone","Please Note",[58,1798,1799],{},"This advisor is not a substitute for professional legal counsel. It offers preliminary guidance on everyday legal questions—helping you decide whether to seek a formal consultation, and how to frame it better when you do.",[54,1801,1803],{"id":1802},"why-this-tool-matters","Why this tool matters",[58,1805,1806,1807,1810],{},"Legal consultation is a standard—and expected—indirect cost for NGOs operating in emergencies. Agencies must follow local law and uphold the rule of law as a matter of principle. This assistant helps you make those consultations count: run your questions through it first to orient yourself, understand the relevant legislation, and arrive at your lawyer's door already knowing what to ask. It's a ",[65,1808,1809],{},"friendly, non-judgmental first stop",", working from an extensive and regularly updated base of Ukrainian legislation.",[54,1812,1814],{"id":1813},"what-it-does-well","What it does well",[58,1816,1817,1818,1821,1822,1825],{},"The assistant answers plain-language questions—in ",[65,1819,1820],{},"English or Ukrainian","—and, crucially, ",[65,1823,1824],{},"backs every answer with the specific laws it relied on",", each cited by name and date so you can check it yourself. It draws on three layers of material, and it keeps them clearly separated:",[490,1827,1828,1834,1840],{},[493,1829,1830,1833],{},[65,1831,1832],{},"Primary legislation"," — the actual laws, Cabinet resolutions, and decrees. This is the authoritative text, and it always comes first.",[493,1835,1836,1839],{},[65,1837,1838],{},"A curated humanitarian knowledge base"," — provisions hand-selected for the realities of NGO work: displacement, martial law, humanitarian imports, taxation, data protection.",[493,1841,1842,1845],{},[65,1843,1844],{},"Expert analysis"," — commentary that interprets or compares the law. This is clearly labelled as analysis, with its date, so it's never mistaken for the law itself.",[54,1847,1849],{"id":1848},"why-its-built-this-way","Why it's built this way",[58,1851,1852,1853,1856],{},"The interesting part of a tool like this isn't the technology—it's the ",[65,1854,1855],{},"discipline"," behind it. A few decisions matter more than the rest, and they're the same decisions I'd defend in any serious AI product:",[58,1858,1859,1862],{},[65,1860,1861],{},"Answers are grounded in real law"," General-purpose chatbots are fluent, confident, and perfectly capable of inventing a law that doesn't exist. For legal questions, that's not a quirk—it's a hazard. So this assistant is deliberately kept on a short leash: it can only answer from legislation it has actually retrieved, and when the source material doesn't cover your question, it says so instead of guessing.",[58,1864,1865,1868],{},[65,1866,1867],{},"Every claim is traceable."," Trust in this domain comes from verifiability, not from a confident tone. That's why each answer points to the exact law behind it, with its title, date, and a link. If it tells you something, you can go and confirm it.",[58,1870,1871,1874,1875,1878],{},[65,1872,1873],{},"The law comes first, and commentary stays labelled."," Mixing an expert's opinion into the statutory text would be misleading. Keeping the primary law, the curated set, and the analysis as distinct, clearly-marked layers means you always know whether you're reading the law or a reading ",[373,1876,1877],{},"of"," the law.",[58,1880,1881,1884],{},[65,1882,1883],{},"It's built for non-lawyers."," The tone is calm and plain-spoken, it works in both English and Ukrainian, and it's honest about its limits—every consequential answer nudges you toward a qualified Ukrainian lawyer for the final word.",[58,1886,1887,1890],{},[65,1888,1889],{},"It stays in its lane."," The assistant is a legal helper and nothing else. It won't be talked into wandering off-topic or into discussing how it was built—it just helps with Ukrainian law, reliably and privately.",[58,1892,1893,1894,1897],{},"None of this makes the tool flashier. It makes it ",[373,1895,1896],{},"trustworthy","—which, for anything that touches compliance and the rule of law, is the only feature that counts.",[54,1899,1901],{"id":1900},"an-evolving-legal-framework","An evolving legal framework",[58,1903,1904,1905,1908,1909,1912],{},"The core law governing humanitarian aid in Ukraine dates back to ",[65,1906,1907],{},"1999"," and has been amended and rephrased many times since February 2022 to reflect wartime conditions. Around it sits a broad body of law covering general economic activity—labour, auditing, imports—plus sector-specific rules, such as those for medical services. Many of these have been profoundly reshaped by ",[65,1910,1911],{},"martial law",", which is exactly why a tool that stays current, cites its sources, and flags martial-law context is so useful.",[54,1914,1916],{"id":1915},"where-it-helps-most","Where it helps most",[58,1918,1919],{},"Based on what's actually keeping teams busy on the ground, these are the areas the assistant is most useful for:",[490,1921,1922,1928,1938,1944,1950,1956,1962,1968],{},[493,1923,1924,1927],{},[65,1925,1926],{},"NGO registration and legal status"," — establishing and maintaining a compliant presence",[493,1929,1930,1933,1934,1937],{},[65,1931,1932],{},"Mobilisation and staff reservation"," (",[373,1935,1936],{},"bronoiuvannia",") — protecting critical national staff",[493,1939,1940,1943],{},[65,1941,1942],{},"Martial law"," — movement restrictions, access, and its effect on operations and contracts",[493,1945,1946,1949],{},[65,1947,1948],{},"Customs and humanitarian imports"," — including the stricter rules around dual-use and medical goods",[493,1951,1952,1955],{},[65,1953,1954],{},"Taxation and financial compliance"," — non-profit status, VAT exemptions on aid, currency controls",[493,1957,1958,1961],{},[65,1959,1960],{},"Data protection"," — handling beneficiary data lawfully, including cross-border transfers",[493,1963,1964,1967],{},[65,1965,1966],{},"Internally displaced persons"," — registration, certification, and access to state support",[493,1969,1970,1973],{},[65,1971,1972],{},"Health operations"," — licensing, medicines, and integration with Ukraine's health systems",[54,1975,1977],{"id":1976},"why-is-this-free","Why is this free?",[58,1979,1980,1981,1984],{},"We have no intention of charging anyone to use it. My work is with ",[65,1982,1983],{},"organizations",", not individual users—I help them build AI products grounded in real operational experience rather than guesswork—so this advisor does two jobs at once.",[58,1986,1987,1988,1991,1992,1995],{},"It's a ",[65,1989,1990],{},"working demonstration"," of what a tailored, domain-specific AI tool can do when it's built carefully around a genuine problem, by someone who has lived that problem. And it's a ",[65,1993,1994],{},"gift to the humanitarian community in Ukraine","—a small contribution to people who spend their days on far harder things than reading legislation.",[58,1997,1998],{},"If it saves a coordinator an hour, or helps a team walk into a legal consultation better prepared, it has done its job.",[1579,2000],{},[58,2002,2003,2004,2007],{},"This advisor exists to make a complex legal environment a little easier to navigate—ultimately ",[65,2005,2006],{},"saving time and resources"," for the organizations doing the work.",[560,2009],{"description":2010,"icon":2011,"link-label":1242,"link-url":565,"title":2012},"Think of it as a conversation with your knowledge base — not a replacement for a lawyer. If your organization would benefit from a private, tailored assistant built around its own documents and workflows, get in touch.","i-ph-chat-circle-text-duotone","Want this for your own documents?",{"title":248,"searchDepth":249,"depth":249,"links":2014},[2015,2016,2017,2018,2019,2020],{"id":1802,"depth":249,"text":1803},{"id":1813,"depth":249,"text":1814},{"id":1848,"depth":249,"text":1849},{"id":1900,"depth":249,"text":1901},{"id":1915,"depth":249,"text":1916},{"id":1976,"depth":249,"text":1977},"2026-07-27T00:00:00.000Z","A free, source-grounded legal assistant for humanitarian action in Ukraine — and why it's built the way it is",[2024],{"label":2025,"url":2026,"icon":2027},"Legal Adviser Demo","\u002Fdemos\u002Fparalegal-advisor","i-ph-scales",{"src":2029,"alt":2030},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FFlux2-Klein_00031_.png","Abstract illustration representing legal documents and AI assistance",{},"\u002Farticles\u002Fparalegal-advisor",{"title":1761,"description":2022},"3.articles\u002F4.paralegal-advisor","7-jbggCwm-6QoygFJPHOEo7brmfzcj_ecMTWij_gow0",{"id":2037,"title":2038,"authors":2039,"badge":2042,"body":2044,"date":2334,"description":2335,"extension":266,"external_links":2336,"image":2341,"meta":2344,"navigation":269,"path":2345,"seo":2346,"stem":2347,"__hash__":2348},"posts\u002F3.articles\u002F5.mapping.md","Mapping with Purpose.",[2040],{"name":44,"to":45,"avatar":2041},{"src":47},{"label":2043},"Mapping",{"type":51,"value":2045,"toc":2329},[2046,2048,2055,2058,2061,2072,2087,2090,2093,2095,2126,2153,2180,2216,2218,2222,2233,2240,2262,2268,2270,2287,2289,2293,2296,2303,2305,2323],[54,2047,1424],{"id":1423},[58,2049,2050,2051,2054],{},"A solid grip on the spatial component is absolutely critical to any humanitarian operation. I'm not talking about abstract jargon like \"humanitarian space\"—I mean the gritty specificity of knowing ",[373,2052,2053],{},"exactly"," where needs are, and where our teams can actually stand on the ground.",[58,2056,2057],{},"Yet, for a tool this critical, the specialized agencies tasked with producing these information management lifelines don't always get it right.",[58,2059,2060],{},"Humanitarian mapping is often too symbolic and too small-scale to be fit for purpose. It is mass-produced to be consumed from comfortable country-level offices.",[58,2062,2063,2064,2067,2068,2071],{},"No wonder the first thing any new Field Coordinator asks when arriving at a staging area is, ",[373,2065,2066],{},"\"Can we make a mapping of presence, guys?\""," The anxiety of planning a mission for a location where needs are already miraculously met (or entirely absent) is very real. The ",[373,2069,2070],{},"5W"," (Who does What, Where, When, and for Whom) was supposed to fill this gap, but the basic information—with correct and functional administrative boundaries—is rarely provided. Strangely, this is incredibly difficult for senior country teams to understand.",[300,2073,2076],{"icon":2074,"title":2075},"i-ph-arrows-in-line-horizontal-duotone","The Core Mapping Paradox: High vs. Low Scale",[58,2077,2078,2079,2082,2083,2086],{},"In cartography, \"scale\" is a fraction. A ",[65,2080,2081],{},"small scale"," map (like 1:1,000,000) shows a large area with very little detail—perfect for a wall poster. A ",[65,2084,2085],{},"large scale"," map (like 1:5,000) zooms in on a tiny area but reveals every building—critical for actual operations. In humanitarian crises, using small scale maps for large scale problems is a recipe for disaster.",[58,2088,2089],{},"An oversimplified map can declare an entire city a \"no-go zone,\" blindly cutting off aid to thousands, while a map that's overly complex can paralyze decision-making with unnecessary noise.",[58,2091,2092],{},"This guide walks you through the hierarchy of mapping. We will explore each level, defining its purpose, audience, and stark limitations. By understanding the difference between a simple sketch and a live geo-server, you'll stop using a hammer when you need a scalpel.",[1579,2094],{},[2096,2097,2104,2117,2120],"level-card",{"icon":2098,"n":2099,"title":2100,"goal":2101,"subtitle":2102,"tool":2103},"i-ph-pencil-line-duotone","1","The Quick Sketch","Immediate communication","Situational Awareness","Drawing software (PowerPoint, Paint, Google My Maps) or simply a pen and paper",[58,2105,2106],{},[82,2107],{"alt":2108,"className":2109,"src":2116},"A quick sketch map demonstrating situational awareness",[2110,2111,2112,2113,2114,86,2115],"w-1\u002F2","sm:w-1\u002F3","float-right","ml-6","mb-4","shadow-md","https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FKH01.png",[58,2118,2119],{},"These are single-use, disposable tools meant to communicate one very specific thing: recent attacks for a donor's visit, a localized office pin, or an emergency hospital route.",[58,2121,2122,2123],{},"This is primarily to show security personnel or VIPs a snapshot of the reality; it proves your situational awareness. They don't need academic analysis—they need an image that will likely get buried and lost in their inbox immediately after they glance at it. ",[65,2124,2125],{},"Do not spend more than 20 minutes on this.",[2096,2127,2134,2141,2144,2147],{"icon":2128,"n":2129,"title":2130,"goal":2131,"subtitle":2132,"tool":2133},"i-ph-presentation-chart-duotone","2","Simple Visualization","Strategic direction","The Storyteller","Desktop GIS export, Canva, Illustrator",[58,2135,2136],{},[82,2137],{"alt":2138,"className":2139,"src":2140},"A simple visualization map demonstrating a strategic direction",[2110,2111,2112,2113,2114,86,2115],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Focha_snapshot_map.png",[58,2142,2143],{},"These are high-scale maps covering vast areas with an extremely low level of detail. Their sheer purpose is to tell a story, serving senior management to gain a vague localization awareness. These are the most abundant in the sector.",[58,2145,2146],{},"This is the typical symbology map. It is useful for a generic planning session at the country office (maybe once a year), and it looks beautiful hanging on the wall to make you feel like a military general strategizing.",[58,2148,2149,2150],{},"In the humanitarian sector, we tend to get fatally stuck in this category, crippling our operational capacities down the line. For a technical coordinator trying to write a shiny proposal, this level is entirely sufficient. They only need to tell the donor: ",[373,2151,2152],{},"\"I want to go here because of the humanitarian situation, but don't worry, I'll figure out the exact locations in the final report.\"",[2096,2154,2161,2168,2171],{"icon":2155,"n":2156,"title":2157,"goal":2158,"subtitle":2159,"tool":2160},"i-ph-chart-line-up-duotone","3","Business Intelligence","Monitoring and inter-regional analysis","The Dashboard","Power BI, Tableau, ActivityInfo, Metabase",[58,2162,2163],{},[82,2164],{"alt":2165,"className":2166,"src":2167},"A dashboard representing business intelligence maps",[2110,2111,2112,2113,2114,86,2115],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fpower_bi_ocha.png",[58,2169,2170],{},"Dashboards represent the next tier of understanding, usually fueled by data collection tools like KoboToolbox and ActivityInfo. This represents a monumental qualitative jump. Before these tools, mountains of data went under-analyzed, and organizations were completely at the mercy of the specific skillset of whoever happened to be the Information Manager that day—ranging from brilliant data scientists to basic Excel users.",[58,2172,2173,2176,2177],{},[65,2174,2175],{},"The Limitation:"," Dashboards shine at middle scales. However, the exact moment you try to force granular, local-level data into them, the sheer volume of information bottlenecks and crashes the software. They simply are not built for microscopic operational tracking. This is a classic example of the humanitarian tech trope: ",[373,2178,2179],{},"\"Keep using it until it breaks.\"",[2096,2181,2188,2195,2202,2205],{"icon":2182,"n":2183,"title":2184,"goal":2185,"subtitle":2186,"tool":2187},"i-ph-map-trifold-duotone","4","Web Map Services","Granular, local queries and field coordination","The Operational Engine","GeoServer, OGC Standards (WMS\u002FWFS)",[58,2189,2190],{},[82,2191],{"alt":2192,"className":2193,"src":2194},"A granular Web Map Service UI",[2110,2111,2112,2113,2114,86,2115],"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002Fua-maps-server.png",[58,2196,2197,2198,2201],{},"These servers handle the absolute granular bedrock of information. If well-configured, they are exactly as good as the detail of the raw information uploaded to them. These are the tools a Field Coordinator ",[373,2199,2200],{},"actually"," needs to decide exactly where to operate.",[58,2203,2204],{},"Because they are perceived as technical, difficult to set up, and because senior management doesn't strictly need them to look good for advocacy, they are consistently overlooked or deemed \"too expensive.\" The truth? Modern open-source backends like GeoServer have completely solved the cost and complexity issues. We just need to implement them.",[58,2206,2207,2208,2215],{},"I deployed a live example of this at ",[65,2209,2210],{},[290,2211,2214],{"href":2212,"rel":2213},"https:\u002F\u002Fua-maps.baena.ai",[294],"ua-maps.baena.ai",". In this service, every single settlement is interactive—clicking on one instantly displays some operational data, such as its exact distance to the frontline. To ensure the context is always reliable, the frontline position and its corresponding buffer zones are automatically recalculated daily. Nevertheless, be mindful when using this service, exact locations are not always updated timely for operational porpuses.",[1579,2217],{},[54,2219,2221],{"id":2220},"the-power-of-scale-the-nikopol-example","The Power of Scale: The Nikopol Example",[58,2223,2224,2225,2228,2229,2232],{},"Understanding the operational difference between ",[65,2226,2227],{},"Level 2"," (Strategic\u002FSmall Scale) and ",[65,2230,2231],{},"Level 4"," (Operational\u002FGranular) isn't an academic debate—it actively saves or ruins lives.",[58,2234,2235,2236,2239],{},"Take the city of ",[65,2237,2238],{},"Nikopol",", which sits under constant artillery threat due to its proximity to Zaporizhzhia Nuclear Power Plant.",[2241,2242,2243,2251],"compare-grid",{},[2244,2245,2248],"compare-card",{"title":2246,"variant":2247},"The Strategic View (Level 2)","bad",[58,2249,2250],{},"On a large-scale map, the whole municipality is painted red to indicate extreme danger. Consequently, it simply falls off the humanitarian planning radar for many agencies because it is lazily deemed \"too complicated to access.\"",[2244,2252,2255],{"title":2253,"variant":2254},"The Operational View (Level 4)","good",[58,2256,2257,2258,2261],{},"A closer look via a granular Web Map Service reveals an entirely different reality. It shows that artillery attacks are heavily concentrated in a narrow strip right along the river coast. ",[373,2259,2260],{},"The rest of the city remains practically untouched",", with an attack ratio no higher than randomly safer cities in the East.",[58,2263,2264,2267],{},[65,2265,2266],{},"The Result:"," By using the wrong scale, we blindly cut off aid to a population. By using the right scale, we pinpoint safe zones for distributions.",[1579,2269],{},[2096,2271,2278,2284],{"icon":2272,"n":2273,"title":2274,"goal":2275,"subtitle":2276,"tool":2277},"i-ph-globe-hemisphere-west-duotone","5","Full GIS & Geoprocessing","Asking complex questions","The Analyst","QGIS, ArcGIS, Python\u002FR",[58,2279,2280,2281],{},"If you need to strictly query your data spatially, you have to lean on raw geoprocessing capabilities. For example: ",[373,2282,2283],{},"\"Exactly how many settlements are within a 20 km distance from the Frontline where combat activity is demonstrably low?\"",[58,2285,2286],{},"You need a full GIS suite to run this complex analysis locally and generate reports on it. This level also includes replicable data pipelines mapping using statistical software (like Python or R), providing the literal foundation for long-term policies and cementing the institutional memory of an operation.",[1579,2288],{},[54,2290,2292],{"id":2291},"conclusion-the-future-is-area-based","Conclusion: The Future is Area-Based",[58,2294,2295],{},"The humanitarian sector desperately needs to force itself up this mapping pyramid. While simple sketch maps have their place, relying exclusively on them deeply hurts our accountability.",[58,2297,2298,2299,2302],{},"The paradigm shift toward ",[65,2300,2301],{},"Area-Based Coordination (ABC)"," will eventually drive this necessary evolution. ABC inherently demands much more refined, generalized, and local data rather than high-level, useless aggregates. This creates the exact feedback loop we need to aggressively promote: better local data enables intelligent field coordination, and superior coordination immediately reveals the gaping need for far more nuanced information systems.",[1579,2304],{},[2096,2306,2310,2320],{"icon":2307,"n":2308,"title":2309},"i-ph-hands-clapping-duotone","0","Bonus: The Foundation",[58,2311,2312,2313,2316,2317],{},"There is actually a ",[65,2314,2315],{},"\"Level 0\""," beneath the pyramid: ",[65,2318,2319],{},"Draw your own maps.",[58,2321,2322],{},"It is a genuinely relaxing experience. When I am trapped in a slow, painfully bureaucratic peace meeting, I color my sketches. Your brain gets infused with the local names, cross-references, and geographical features, quietly building a deeply intuitive understanding of the terrain that software alone can simply never replace.",[560,2324],{"description":2325,"icon":2326,"link-label":2327,"link-url":2212,"title":2328},"Do you need a granular, scale-appropriate mapping solution for your field operations? Let's build something that works.","i-heroicons-map","Explore ua-maps.baena.ai","Ready to upgrade your operational mapping?",{"title":248,"searchDepth":249,"depth":249,"links":2330},[2331,2332,2333],{"id":1423,"depth":249,"text":1424},{"id":2220,"depth":249,"text":2221},{"id":2291,"depth":249,"text":2292},"2018-05-15T00:00:00.000Z","An opinionated guide for choosing the right visual and its mapping tool.",[2337],{"label":2338,"url":2339,"icon":2340},"Ukraine Mapping Service","\u002Fdemos\u002Fukraine-mapping-service","i-ph-map-pin",{"src":2342,"alt":2343},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00164_.png","Conceptual illustration of humanitarian mapping at different operational scales",{},"\u002Farticles\u002Fmapping",{"title":2038,"description":2335},"3.articles\u002F5.mapping","OX49_T9qUqp9xnj8EGXcNYPW5uDCTtq0ofGYZB3ZttI",{"id":2350,"title":13,"authors":2351,"badge":2354,"body":2356,"date":2410,"description":2411,"extension":266,"external_links":2412,"image":2420,"meta":2423,"navigation":269,"path":2424,"seo":2425,"stem":2426,"__hash__":2427},"posts\u002F3.articles\u002F6.mpca-ukraine.md",[2352],{"name":44,"to":45,"avatar":2353},{"src":47},{"label":2355},"MPCA | Information Management",{"type":51,"value":2357,"toc":2405},[2358,2361,2364,2368,2371,2374,2378,2389,2392,2396,2399,2402],[58,2359,2360],{},"In response to the conflict in the Donbas region and the broader war, Multipurpose Cash Assistance (MPCA) has emerged as a cornerstone of the humanitarian response in Ukraine. By providing flexible financial support, MPCA empowers affected households to prioritize their own diverse needs.",[58,2362,2363],{},"To understand the effectiveness and reach of this critical intervention, I conducted a study utilizing data collected by the United Nations Office for the Coordination of Humanitarian Affairs (OCHA). Regularly updated by field reports, this dataset documents interventions across the country, offering a window into funding allocations, demographic reach, and regional coverage.",[54,2365,2367],{"id":2366},"the-tension-between-access-and-need","The Tension Between Access and Need",[58,2369,2370],{},"One of the most significant insights from the data was the visible influence of logistics on aid distribution. We observed a significant clustering of activities in regions further from direct conflict zones. This suggests that while needs are acute near the front lines, logistical ease and established organizational presence often drive where aid is delivered—a finding that highlights the constant challenge of balancing safety with purely needs-based criteria.",[58,2372,2373],{},"Despite these logistical hurdles, the data showed a commendable commitment to rural populations. NGOs and the Red Cross\u002FRed Crescent Movement, in particular, demonstrated a strong focus on reaching these often-overlooked areas, ensuring that support wasn't strictly confined to major urban centers.",[54,2375,2377],{"id":2376},"granularity-and-the-invisible-demographics","Granularity and the \"Invisible\" Demographics",[58,2379,2380,2381,2384,2385,2388],{},"When analyzing ",[373,2382,2383],{},"who"," is reporting the data, a clear divergence emerged. National NGOs excelled in providing detailed, local-level reporting, offering high-definition visibility into their operations. mainly, UN agencies tended to report at the broader ",[373,2386,2387],{},"Raion"," level, which, while efficient for macro-reporting, creates challenges when trying to identify precise gaps in service delivery.",[58,2390,2391],{},"Furthermore, the demographic breakdowns revealed potential blind spots. There appears to be an underreporting of adult men in household counts. This statistical anomaly warrants further investigation, as it could impact the calculation of aid adequacy for families, potentially leaving specific vulnerabilities unaddressed.",[54,2393,2395],{"id":2394},"turning-coordination-data-into-analytical-insight","Turning Coordination Data into Analytical Insight",[58,2397,2398],{},"While this dataset is primarily designed for coordination, this analysis proves it holds immense value for deeper research—provided we navigate its limitations carefully. Inconsistencies, particularly in financial reporting, highlight the need for careful interpretation and triangulation with other sources. However, the dataset's strength lies in its rich geographical detail and the transparency of implementing partners.",[58,2400,2401],{},"Looking forward, this analysis supports the recommendations of the MPCA technical working group. By standardizing financial reporting and refining demographic breakdowns, we can significantly improve future tracking tools—not just for Ukraine, but for humanitarian operations globally.",[58,2403,2404],{},"This study has served as a launchpad for more advanced inquiries. It has paved the way for my current work on using statistical models to understand risk perception, demonstrating how data gathered for coordination can be transformed into strategic foresight.",{"title":248,"searchDepth":249,"depth":249,"links":2406},[2407,2408,2409],{"id":2366,"depth":249,"text":2367},{"id":2376,"depth":249,"text":2377},{"id":2394,"depth":249,"text":2395},"2024-02-01T00:00:00.000Z","An analysis of Multipurpose Cash Assistance (MPCA) data in Ukraine, highlighting key findings and implications for humanitarian aid.",[2413,2416],{"label":2414,"url":2415,"icon":1747},"MPCA Dashboard","\u002Fdemos\u002Fmpca-ukraine-report",{"label":2417,"url":2418,"icon":2419},"Full Interactive Report","https:\u002F\u002Fstatic-reports.baena.info\u002Freports\u002F241101-mpca-in-ukraine\u002Findex.html","i-ph-file-text",{"src":2421,"alt":2422},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FFlux2-Klein_00026_.png","MPCA Analysis",{},"\u002Farticles\u002Fmpca-ukraine",{"title":13,"description":2411},"3.articles\u002F6.mpca-ukraine","FTN4NMUrhxYnmRNBGJyBPIUX4pQJJZzGxm3GJ8rGOVw",{"id":2429,"title":2430,"authors":2431,"badge":2434,"body":2436,"date":2673,"description":2674,"extension":266,"external_links":2675,"image":2681,"meta":2684,"navigation":269,"path":2685,"seo":2686,"stem":2687,"__hash__":2688},"posts\u002F3.articles\u002F7.uhf-twin-tiers.md","The Ceiling-Bound Economy of the Ukraine Humanitarian Fund",[2432],{"name":44,"to":45,"avatar":2433},{"src":47},{"label":2435},"Funding Analysis | CBPF",{"type":51,"value":2437,"toc":2661},[2438,2442,2457,2460,2463,2467,2478,2485,2489,2492,2506,2513,2517,2520,2523,2529,2536,2540,2547,2554,2558,2561,2567,2577,2583,2589,2592,2596,2599,2610,2613,2617,2620,2623,2627,2630,2635,2639,2642,2645,2652,2655,2657],[54,2439,2441],{"id":2440},"how-a-standard-shape-cbpf-with-a-2x-envelope-produces-a-ceiling-bound-grant-economy","How a standard-shape CBPF with a 2x envelope produces a ceiling-bound grant economy",[300,2443,2446],{"icon":2444,"title":2445},"i-ph-file-text-duotone","About This Piece",[58,2447,2448,2451,2452,2456],{},[65,2449,2450],{},"Full technical report:"," ",[290,2453,2455],{"href":2454},"\u002Fdemos\u002Fukraine-poolfund-report","available here",". This is an explanatory, opinionated companion piece — views expressed are solely my own and do not reflect the positions of any organisation I have worked with.",[58,2458,2459],{},"On the usual concentration metrics, Herfindahl index, top-5 share, top-10 share, the Ukraine Humanitarian Fund (UHF) looks like a fairly ordinary Country-Based Pooled Fund. That is exactly why it is misunderstood. The UHF is not more concentrated than its peers; it is much larger. And when you run a normal-shape distribution through a $541M envelope over three years, the arithmetic produces something peers don't have: a cap-binding upper tier that receives the same maximum grant, round after round, year after year.",[58,2461,2462],{},"The result is effectively a two-tier fund hiding inside a one-number summary.",[54,2464,2466],{"id":2465},"the-inner-circle-a-cap-binding-core","The inner circle: a cap-binding core",[58,2468,2469,2470,2473,2474,2477],{},"Twenty organisations appeared in every UHF allocation round between 2022 and 2024. Together they absorbed ",[65,2471,2472],{},"$299M of the $541M envelope, about 55%",". That alone would be unremarkable for a mega-fund. What makes the UHF distinctive is what happens at the very top of this cohort: seven partners, ACTED, UNHCR, DRC, People in Need, ICF Caritas Ukraine, NRC and Proliska, average grants of ",[65,2475,2476],{},"$3.4M-$4.8M",", essentially every ticket at or near the $5M standard cap. Six of the seven have also received at least one exceptional-cap grant above $6M (UNHCR's highest: $9.4M).",[58,2479,2480,2481,2484],{},"This is not a concentration story in the classical sense. It is a ",[65,2482,2483],{},"ceiling-binding"," story. For these partners, the binding constraint is no longer competition or absorptive capacity, it is the programmatic cap itself.",[54,2486,2488],{"id":2487},"the-flat-top-a-distribution-peers-dont-have","The flat top: a distribution peers don't have",[58,2490,2491],{},"The clearest fingerprint of the UHF's scale is the shape of its grant distribution at the top.",[490,2493,2494,2501],{},[493,2495,2496,2497,2500],{},"The ",[65,2498,2499],{},"median UHF grant is 2.6x the global CBPF median",": $1.13M vs $440K.",[493,2502,2496,2503],{},[65,2504,2505],{},"90th percentile is exactly $5.00M, the standard cap.",[58,2507,2508,2509,2512],{},"That second number is unique among CBPFs. Peer funds, Afghanistan, Yemen, Ethiopia, Somalia, oPt, show a long right tail that fades out smoothly as grants get larger. Ukraine shows a ",[65,2510,2511],{},"flat top",": a dense cluster of grants pinned against the ceiling. The cap is doing visible work in the data.",[54,2514,2516],{"id":2515},"the-disguise-why-the-averages-look-normal","The disguise: why the averages look normal",[58,2518,2519],{},"Given that flat top, one might expect the UHF's concentration metrics to scream. They don't. HHI, top-5 and top-10 shares all place Ukraine mid-pack among broad-partnership CBPFs. The paradox dissolves once you look past the Core-20.",[58,2521,2522],{},"The other 74 partners in the fund's 2022-2024 roster share the remaining $242M, 45% of the envelope, at much smaller, less repeated tickets. A churning outer ring of single- and two-grant partners operates at a fraction of the core's scale and drags the average down.",[58,2524,2525,2526],{},"When you blend a tightly capped top tier with a long, flat bottom tier, the aggregate metrics land squarely in normal CBPF territory. ",[65,2527,2528],{},"Averages here actively obscure the operational reality.",[58,2530,2531,2532,2535],{},"The honest summary is: ",[373,2533,2534],{},"same shape as peers x roughly double the envelope = roughly double the mean partner take",". Everything unusual about Ukraine grant sizes follows mechanically from that arithmetic.",[54,2537,2539],{"id":2538},"the-un-exit-and-the-localized-vacuum","The UN Exit and the localized vacuum",[58,2541,2542,2543,2546],{},"Where did this excess volume for NGOs come from? The data reveals a dramatic, structural policy shift: a ",[65,2544,2545],{},"92% collapse in UN Agency funding"," over just two years. UN allocations plummeted from $66M in 2022 to a mere $5M in 2024.",[58,2548,2549,2550,2553],{},"This vacuum didn't shrink the fund; it redistributed the wealth. The displaced volume flowed directly into a broadened National NGO tier and select INGOs (like DRC and Caritas Ukraine). This created an unprecedented dynamic where, by 2024, the median grant awarded to a Ukrainian NNGO was actually ",[373,2551,2552],{},"larger"," than the median grant awarded to an INGO—an inversion of the global norm.",[54,2555,2557],{"id":2556},"what-sits-inside-each-tier-a-risk-weighted-look-at-the-recipients","What sits inside each tier: a risk-weighted look at the recipients",[58,2559,2560],{},"Not every recipient carries the same due-diligence profile. Four clusters are worth naming explicitly.",[58,2562,2563,2566],{},[65,2564,2565],{},"(a) \"Born of the war\" partners, no pre-invasion track record."," A handful of organisations were established in direct response to the 2022 escalation. When these entities receive multi-million-dollar grants, their institutional existence is effectively endogenous to UHF-scale funding. There is no pre-war baseline against which to assess organisational risk.",[58,2568,2569,2572,2573,2576],{},[65,2570,2571],{},"(b) Small global revenue base, UHF volume dwarfs the institution."," Some international partners operated with pre-war global revenues of a few million dollars. A single UHF grant can equal or exceed their entire historical annual turnover, structurally transforming the organisation. The technical report tracks this as a ",[65,2574,2575],{},"dependency ratio",": UHF disbursements divided by 2021 global revenue. Several partners exceed 1.0; one exceeds 4x.",[58,2578,2579,2582],{},[65,2580,2581],{},"(c) Established local NNGOs, the localisation success story."," A cohort of Ukrainian NGOs with 5-10 years of pre-2022 operational history was absorbed into the UHF ecosystem as part of the deliberate post-invasion localisation shift. (Per UHF Annual Reports, direct allocations to national partners moved from ~30% pre-war toward ~60% post-invasion.) These are the partners the fund can legitimately point to when describing its localisation outcomes.",[58,2584,2585,2588],{},[65,2586,2587],{},"(d) The hyperscalers."," Major UN agencies and Tier-1 INGOs with the global infrastructure to absorb rapid funding surges without altering their institutional risk profile. These sit predominantly inside the seven cap-binding partners.",[58,2590,2591],{},"The interesting policy question is not how to shrink the inner circle. It is how thin the empirical basis is for the partners in category (a) and (b) compared to (c) and (d), and whether that thinness is being priced into risk management.",[54,2593,2595],{"id":2594},"the-ideal-candidate-the-full-coverage-premium","The Ideal Candidate: The Full-Coverage Premium",[58,2597,2598],{},"If we map every UHF partner since 2019 across two simple axes—geographic footprint (oblasts) and sector coverage (clusters)—a clear \"Ideal Candidate\" profile emerges.",[58,2600,2601,2602,2605,2606,2609],{},"Organisations that operate across multiple oblasts and sectors, the ",[65,2603,2604],{},"Full-Coverage partners",", capture the vast majority of the disbursed budget. Only 16% of all partners sit in this top quadrant, but they absorbed ",[65,2607,2608],{},"46% of all UHF disbursements since 2019"," (a median cumulative budget of $20.9M per partner). Conversely, narrow Specialists have a median budget of just $2.2M.",[58,2611,2612],{},"The path to scale is brutally clear: Geographic and sector coverage are the strongest predictors of funding volume.",[54,2614,2616],{"id":2615},"health-fragmentation-over-expertise","Health: Fragmentation over Expertise",[58,2618,2619],{},"This absorption-first model creates blind spots, nowhere more visible than the Health sector. Despite massive allocations, the UHF’s health portfolio has seen profound temporal fragmentation. As institutional anchor agencies like the WHO exited the direct recipient pool, the funding flowed to a shifting mix of INGOs and NNGOs selected for their grant-absorption capacity, rather than technical medical expertise or pre-war field presence.",[58,2621,2622],{},"Even more glaring: the strategy systematically overlooked Local Health Facilities as partnership candidates—concentrating resources in predominantly Anglo-Saxon NGOs lacking prior health footprints in Ukraine, while ignoring the existing domestic healthcare infrastructure.",[54,2624,2626],{"id":2625},"interactive-grant-size-distribution","Interactive Grant Size Distribution",[58,2628,2629],{},"Explore the grant distribution data interactively:",[2631,2632],"fullscreen-embed",{"src":2633,"title":2634},"https:\u002F\u002Fshiny.baena.info\u002Fuhf-grant-size-2025\u002F","UHF Grant Size Distribution",[54,2636,2638],{"id":2637},"conclusion-a-dual-track-response-and-the-reality-for-applicants","Conclusion: a dual-track response, and the reality for applicants",[58,2640,2641],{},"The UHF is a unique operational model in the CBPF system: not one fund, but two operating under identical headline statistics. There is an inner tier of twenty partners recycling maximum ceiling grants round after round, shielded by their massive absorption capacity and broad coverage footprint. Then, a churning outer ring of 74 smaller-ticket partners that make the aggregate fund metrics look \"normal.\"",[58,2643,2644],{},"For organisations deciding whether to invest in a UHF proposal, the implication is harsh: the defining constraint is not the quality of your pitch—it is the shape of the grant distribution itself.",[58,2646,2647,2648,2651],{},"If your institutional profile (pre-war track record, geographic coverage, sectoral breadth) doesn't mirror the Core-20's ",[65,2649,2650],{},"Full-Coverage"," model, the standard allocation is unlikely to elevate your rank, no matter how robust the proposal. You are competing against an arithmetic inevitability—a $5M ceiling on a $540M envelope that guarantees large repeated grants to those with proven scale.",[58,2653,2654],{},"For the fund itself, the hardest question is whether a flat-topped, cap-binding distribution that increasingly ignores local health facilities and over-relies on Reserve Allocations (51.9% globally vs the 40.2% CBPF average) is what its donors and the people it serves actually need. Is this deliberate strategic allocation, or simply what happens when you run the standard playbook at double the scale?",[1579,2656],{},[560,2658],{"description":2659,"icon":2155,"link-label":1242,"link-url":565,"title":2660},"I offer task-based consultancy for organisations that want this kind of analysis done on their own data.","Need a similar report on your own data?",{"title":248,"searchDepth":249,"depth":249,"links":2662},[2663,2664,2665,2666,2667,2668,2669,2670,2671,2672],{"id":2440,"depth":249,"text":2441},{"id":2465,"depth":249,"text":2466},{"id":2487,"depth":249,"text":2488},{"id":2515,"depth":249,"text":2516},{"id":2538,"depth":249,"text":2539},{"id":2556,"depth":249,"text":2557},{"id":2594,"depth":249,"text":2595},{"id":2615,"depth":249,"text":2616},{"id":2625,"depth":249,"text":2626},{"id":2637,"depth":249,"text":2638},"2026-04-23T00:00:00.000Z","Why geographic footprint and absorption capacity—not proposal quality—make or break your funding.",[2676,2678],{"label":2677,"url":2454,"icon":2419},"Full Technical Report",{"label":1749,"url":2679,"icon":2680},"https:\u002F\u002Fgithub.com\u002FJesus-Baena\u002F2026-ukraine-poolfund-analysis","i-ph-github-logo",{"src":2682,"alt":2683},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00518_.png","Ukraine Humanitarian Fund analysis illustration",{},"\u002Farticles\u002Fuhf-twin-tiers",{"title":2430,"description":2674},"3.articles\u002F7.uhf-twin-tiers","-mKE4iAa5HztVcJ8vfKJmQefYvH5yHQLxK92gh-pEo8",{"id":2690,"title":2691,"authors":2692,"badge":2695,"body":2697,"date":2021,"description":2986,"extension":266,"external_links":2987,"image":2992,"meta":2994,"navigation":269,"path":2995,"seo":2996,"stem":2997,"__hash__":2998},"posts\u002F3.articles\u002F8.paralegal-advisor-engineering.md","Engineering a Trustworthy Legal RAG",[2693],{"name":44,"to":45,"avatar":2694},{"src":47},{"label":2696},"AI | Engineering",{"type":51,"value":2698,"toc":2968},[2699,2706,2713,2717,2720,2758,2762,2772,2782,2786,2790,2801,2805,2816,2820,2823,2827,2830,2853,2860,2864,2871,2875,2882,2886,2893,2897,2900,2920,2924,2935,2939,2946,2950,2953,2957,2964],[58,2700,2701,2702,2705],{},"This is the technical companion to ",[290,2703,2704],{"href":2032},"the overview of the Paralegal Advisor",". The overview explains what the tool does and why it's free; this piece is for people who build these systems and want to understand the decisions underneath it.",[58,2707,2708,2709,2712],{},"The short version: ",[65,2710,2711],{},"for high-stakes, domain-specific AI, the hard part isn't the model — it's the discipline around it."," A general chatbot that sounds like a lawyer is easy. One you can trust with a compliance question is not. Almost every decision below is in service of that difference.",[54,2714,2716],{"id":2715},"the-constraints-that-shaped-everything","The constraints that shaped everything",[58,2718,2719],{},"Before any architecture, the problem set the rules:",[490,2721,2722,2728,2734,2740,2746,2752],{},[493,2723,2724,2727],{},[65,2725,2726],{},"Wrong answers are expensive."," This touches compliance and the rule of law. A confident hallucination isn't a bug, it's a liability.",[493,2729,2730,2733],{},[65,2731,2732],{},"The law moves."," Ukraine's humanitarian legal framework has been amended continuously since February 2022. The corpus is a moving target.",[493,2735,2736,2739],{},[65,2737,2738],{},"Two languages."," Source texts are Ukrainian; many users work in English.",[493,2741,2742,2745],{},[65,2743,2744],{},"Non-expert users."," Coordinators, not lawyers.",[493,2747,2748,2751],{},[65,2749,2750],{},"It has to be verifiable."," Users must be able to check what they're told.",[493,2753,2754,2757],{},[65,2755,2756],{},"It has to be cheap and durable."," It's a free public tool, so runaway cost or fragility would kill it.",[54,2759,2761],{"id":2760},"architecture-at-a-glance","Architecture at a glance",[2763,2764,2769],"pre",{"className":2765,"code":2767,"language":2768},[2766],"language-text","INGEST                         SERVE\nofficial legislation           user question\n  → extract & normalize          → embed the query\n  → chunk                        → retrieve across 3 corpora (law-first)\n  → embed (Gemini)               → reason over sources (tool agent)\n  → vector store (Qdrant)        → answer + cited sources\n        ▲                              │\n        └── incremental updates        └── same-origin proxy → chat UI\n","text",[2770,2771,2767],"code",{"__ignoreMap":248},[58,2773,2774,2775,2777,2778,2781],{},"The stack is deliberately boring: ",[65,2776,1385],{}," for embeddings and generation, ",[65,2779,2780],{},"Qdrant"," as the vector store, a low-code orchestration layer for the serving graph, and a thin same-origin web layer. Boring is a feature — the value is in the choices, not the novelty.",[54,2783,2785],{"id":2784},"the-decisions-that-matter","The decisions that matter",[167,2787,2789],{"id":2788},"retrieval-not-fine-tuning","Retrieval, not fine-tuning",[58,2791,2792,2793,2796,2797,2800],{},"The corpus changes every week. Fine-tuning a model on legislation would mean retraining every time a law is amended, and it would still blur the line between \"the model learned this\" and \"the model is guessing.\" ",[65,2794,2795],{},"Retrieval-Augmented Generation"," keeps the knowledge in a store you can update in minutes and inspect directly, and it makes citation natural: the model answers ",[373,2798,2799],{},"from"," passages it just retrieved. For a domain where provenance is the whole point, RAG isn't a shortcut — it's the correct architecture.",[167,2802,2804],{"id":2803},"embeddings-quality-and-language-first","Embeddings: quality and language first",[58,2806,2807,2808,2811,2812,2815],{},"The corpus was migrated to Gemini's embedding model at its full dimensionality. Two reasons drove it. First, ",[65,2809,2810],{},"multilingual quality"," — the same model has to place a Ukrainian statute and an English question near each other in vector space. Second, ",[65,2813,2814],{},"query\u002Fdocument asymmetry",": questions and legal passages are embedded with different task types, so a short question retrieves long dense text well. These are the kinds of details that quietly determine whether retrieval works at all.",[167,2817,2819],{"id":2818},"chunking-sized-to-the-law-not-to-a-default","Chunking sized to the law, not to a default",[58,2821,2822],{},"Legal text has natural units — articles and sections. Chunk too small (a habit inherited from older, smaller embedding models) and you shred an article across fragments that each lose the thread. The chunk size here is tuned so that most individual provisions stay whole while comfortably fitting the embedding model's input budget. Chunking is unglamorous and one of the highest-leverage decisions in any RAG system.",[167,2824,2826],{"id":2825},"three-corpora-law-first","Three corpora, law-first",[58,2828,2829],{},"Not all sources carry equal authority, and mixing them would be misleading. The knowledge base is split into three:",[2831,2832,2833,2838,2844],"ol",{},[493,2834,2835,2837],{},[65,2836,1832],{}," — the authoritative statutory text.",[493,2839,2840,2843],{},[65,2841,2842],{},"A curated humanitarian set"," — provisions hand-selected for NGO operations, enriched with relevance metadata.",[493,2845,2846,2849,2850,1878],{},[65,2847,2848],{},"Secondary analysis"," — expert commentary that ",[373,2851,2852],{},"reviews",[58,2854,2855,2856,2859],{},"The system always leads with primary law and treats analysis as clearly-labelled commentary, tagged with its date. Encoding the ",[65,2857,2858],{},"authority hierarchy"," into the retrieval layer — rather than hoping the model infers it — is what stops \"an expert's opinion\" from being served as \"the law.\"",[167,2861,2863],{"id":2862},"a-tool-agent-not-a-single-retriever-chain","A tool agent, not a single-retriever chain",[58,2865,2866,2867,2870],{},"The obvious first build is one retriever feeding one prompt. It can't serve three corpora with different roles. So the serving graph is a ",[65,2868,2869],{},"tool-using agent",": the model has a separate retrieval tool per corpus and decides which to consult, always starting with primary law. The trade-off is honest — several model calls per question instead of one, so a few extra seconds of latency. In exchange you get source separation, law-first behaviour, and the ability to say \"the sources don't cover this\" instead of forcing an answer. For this domain, that trade is worth it.",[167,2872,2874],{"id":2873},"metadata-dates-and-ranking","Metadata, dates, and ranking",[58,2876,2877,2878,2881],{},"Every chunk carries structured metadata: area of law, enactment\u002Famendment date, and (for the curated set) an operational-relevance score. Dates are normalized to a sortable form so the store can do real range queries — \"only laws since X\" — rather than string matching. Ranking by operational relevance means the corpus reflects ",[65,2879,2880],{},"field experience",", not just whatever the scraper found first.",[167,2883,2885],{"id":2884},"citations-as-a-first-class-output","Citations as a first-class output",[58,2887,2888,2889,2892],{},"Every answer returns the specific sources behind it — title, date, link — rendered as numbered cards, with analysis visually distinguished from statute. This isn't decoration. In a trust-sensitive domain, ",[65,2890,2891],{},"verifiability is the product","; a citation the user can open is worth more than any amount of confident phrasing.",[167,2894,2896],{"id":2895},"guardrails-by-design","Guardrails, by design",[58,2898,2899],{},"Three families of guardrail, all specified rather than hoped for:",[490,2901,2902,2908,2914],{},[493,2903,2904,2907],{},[65,2905,2906],{},"Grounding."," The assistant answers from retrieved text and is instructed to say when the material doesn't cover a question — never to invent a law, article, or date.",[493,2909,2910,2913],{},[65,2911,2912],{},"Humility."," Consequential answers point the user to a qualified lawyer. It's a first stop, not a substitute.",[493,2915,2916,2919],{},[65,2917,2918],{},"Boundaries."," It stays a legal assistant: it won't be talked off-topic, and it won't disclose its own internals. Guardrails you can't state precisely are guardrails you don't really have.",[167,2921,2923],{"id":2922},"serving-and-access","Serving and access",[58,2925,2926,2927,2930,2931,2934],{},"The web layer never sees an API key — requests are proxied ",[65,2928,2929],{},"same-origin",", with the credential injected server-side. Access is then shaped by layered controls: an origin check to stop other sites embedding the tool, plus rate limiting as a fair-use and cost ceiling. The principle worth stating: an origin check is a ",[373,2932,2933],{},"browser-scope"," control, so it's treated as one layer among several, not as authentication. Defense in depth, appropriate to a free public demo.",[167,2936,2938],{"id":2937},"cost-and-durability","Cost and durability",[58,2940,2941,2942,2945],{},"A free tool has to stay free. That means a cost-efficient generation model, streaming responses so the interface feels quick even while the agent works, and a spending ceiling so a traffic spike can't turn into a surprise bill. Keeping the corpus current is handled by incremental updates and, when the embedding model itself changes, a ",[65,2943,2944],{},"zero-downtime rebuild"," into a fresh collection before switching over — plus quality gates on extraction so malformed documents never reach the store.",[54,2947,2949],{"id":2948},"honest-trade-offs","Honest trade-offs",[58,2951,2952],{},"No system is finished, and pretending otherwise is a red flag. A few things I'd flag openly: the agent's multi-call design trades latency for quality; translation quality for citation titles is good but not a substitute for a native reading of the statute; and a proper offline evaluation harness — measuring retrieval precision and answer faithfulness against a labelled question set — is the natural next investment for a system like this.",[54,2954,2956],{"id":2955},"why-write-this-up","Why write this up",[58,2958,2959,2960,2963],{},"The tools here are commodities; anyone can wire them together. What makes a domain AI product trustworthy is the ",[65,2961,2962],{},"judgment"," — knowing that chunking matters more than model choice, that authority hierarchy belongs in the retrieval layer, that verifiability beats fluency, and that guardrails have to be specified, not assumed. That judgment comes from having lived the problem, not just read about it.",[560,2965],{"description":2966,"icon":563,"link-label":564,"link-url":565,"title":2967},"That's exactly the kind of work I do with organizations: taking a real operational problem and building an AI product that's accurate, verifiable, and durable enough to rely on.","Building something similar?",{"title":248,"searchDepth":249,"depth":249,"links":2969},[2970,2971,2972,2984,2985],{"id":2715,"depth":249,"text":2716},{"id":2760,"depth":249,"text":2761},{"id":2784,"depth":249,"text":2785,"children":2973},[2974,2975,2976,2977,2978,2979,2980,2981,2982,2983],{"id":2788,"depth":255,"text":2789},{"id":2803,"depth":255,"text":2804},{"id":2818,"depth":255,"text":2819},{"id":2825,"depth":255,"text":2826},{"id":2862,"depth":255,"text":2863},{"id":2873,"depth":255,"text":2874},{"id":2884,"depth":255,"text":2885},{"id":2895,"depth":255,"text":2896},{"id":2922,"depth":255,"text":2923},{"id":2937,"depth":255,"text":2938},{"id":2948,"depth":249,"text":2949},{"id":2955,"depth":249,"text":2956},"The design decisions behind the Ukraine Paralegal Advisor — and why they're the right ones for high-stakes, domain-specific AI",[2988,2989],{"label":2025,"url":2026,"icon":2027},{"label":2990,"url":2032,"icon":2991},"Read the overview","i-ph-article",{"src":2029,"alt":2993},"Abstract illustration of legal documents and retrieval-augmented AI",{},"\u002Farticles\u002Fparalegal-advisor-engineering",{"title":2691,"description":2986},"3.articles\u002F8.paralegal-advisor-engineering","19IWoIhCbs-Gl2Ss5NwxVLUQ7wuPOvdnjQ9CqLEH9qM",{"id":3000,"title":3001,"authors":3002,"badge":3005,"body":3007,"date":3497,"description":3498,"extension":266,"external_links":27,"image":3499,"meta":3502,"navigation":269,"path":3503,"seo":3504,"stem":3505,"__hash__":3506},"posts\u002F3.articles\u002F9.multilingual-humanitarian-rag.md","Adding Bangla, Swahili and Hausa to a sovereign humanitarian assistant",[3003],{"name":44,"to":45,"avatar":3004},{"src":47},{"label":3006},"AI | Evaluation",{"type":51,"value":3008,"toc":3487},[3009,3012,3019,3022,3026,3029,3059,3065,3069,3072,3075,3109,3113,3120,3198,3205,3209,3216,3274,3277,3280,3298,3305,3309,3312,3315,3342,3349,3353,3360,3367,3371,3378,3417,3423,3438,3442,3449,3456,3473,3476,3481,3484],[58,3010,3011],{},"Our humanitarian assistant answers questions about the normative canon of the sector — the Sphere Handbook, the Geneva Conventions, the 1951 Refugee Convention and their siblings — from a corpus we ingest and embed ourselves, behind an egress boundary where every model call is logged, redacted and attributable. The corpus is English. The people who need it often aren't English speakers.",[58,3013,3014,3015,3018],{},"So we decided to add three languages: ",[65,3016,3017],{},"Bangla, Swahili and Hausa",". Together they cover humanitarian contexts from Cox's Bazar to the Sahel. Individually, they are three very different engineering problems — which we only know because we measured each one before building anything.",[58,3020,3021],{},"This is the story of those measurements: what they cost (single-digit euros), what they found, and the working method that made the architecture emerge from the data rather than from a preconceived opinion.",[54,3023,3025],{"id":3024},"four-layers-four-ways-to-fail","Four layers, four ways to fail",[58,3027,3028],{},"A language isn't \"supported\" because the model card says so. For a user writing in Bangla, four independent layers have to work:",[2831,3030,3031,3037,3047,3053],{},[493,3032,3033,3036],{},[65,3034,3035],{},"Detect"," — know which language the message is in. Our existing heuristic, built for a Ukrainian\u002FRussian\u002FEnglish world, classified all three new languages as English.",[493,3038,3039,3042,3043,3046],{},[65,3040,3041],{},"Retrieve"," — find the right ",[373,3044,3045],{},"English"," passages from a query written in another language. Cross-lingual retrieval is the embedding model's job.",[493,3048,3049,3052],{},[65,3050,3051],{},"Generate"," — answer in the user's language, grounded in English evidence, without reaching for a provider that breaks the sovereignty posture.",[493,3054,3055,3058],{},[65,3056,3057],{},"Redact"," — catch names, phone numbers and national IDs before anything crosses the perimeter. Fail closed, per language.",[58,3060,3061,3062,320],{},"Each layer can fail on its own, but it happens that ",[65,3063,3064],{},"they fail differently per language",[54,3066,3068],{"id":3067},"measure-first-rent-nothing","Measure first, rent nothing",[58,3070,3071],{},"Before committing to any architecture (or any GPU), we ran two probes on a serverless, per-token EU inference tier. About 2,900 model calls, less than five euros, no instance ever provisioned.",[58,3073,3074],{},"The probes inherited the evaluation discipline we'd already paid tuition for on the English stack:",[490,3076,3077,3083,3089,3103],{},[493,3078,3079,3082],{},[65,3080,3081],{},"Repeat everything."," Our eval has a known noise floor (~4 points on a 100-point scale between identical runs). Any single-run comparison smaller than that is measuring the generator's mood. Every number below is a mean of three runs, with the spread reported.",[493,3084,3085,3088],{},[65,3086,3087],{},"Score retrieval separately from generation."," Doing it together, it prevents observability.",[493,3090,3091,3094,3095,3098,3099,3102],{},[65,3092,3093],{},"Verify your translations."," The eval questions were machine-translated, then round-trip-checked by an ",[373,3096,3097],{},"independent"," model with a third model judging semantic equivalence. Survival rates: Bangla 32\u002F40, Swahili 25\u002F40, Hausa 22\u002F40 — and one Bangla flag was a silent measles–cholera substitution. That's a finding in itself: machine-translated ",[373,3100,3101],{},"content"," needs review, whatever you do about conversation.",[493,3104,3105,3108],{},[65,3106,3107],{},"Distrust your judge."," The Hausa results were re-judged in full by a second model from a different family. The ranking survived; the absolute numbers moved. Report both.",[54,3110,3112],{"id":3111},"finding-1-generation-one-model-is-language-uniform","Finding 1 — Generation: one model is language-uniform",[58,3114,3115,3116,3119],{},"We gave three candidate models the ",[373,3117,3118],{},"correct"," English evidence and the question in each language, requiring the answer in that language. This isolates cross-lingual generation from retrieval entirely.",[617,3121,3122,3140],{},[620,3123,3124],{},[623,3125,3126,3129,3131,3134,3137],{},[626,3127,3128],{},"Model",[626,3130,3045],{},[626,3132,3133],{},"Bangla",[626,3135,3136],{},"Swahili",[626,3138,3139],{},"Hausa",[636,3141,3142,3163,3180],{},[623,3143,3144,3147,3150,3155,3158],{},[641,3145,3146],{},"gemma-4-26b",[641,3148,3149],{},"0.79",[641,3151,3152],{},[65,3153,3154],{},"0.88",[641,3156,3157],{},"0.81",[641,3159,3160],{},[65,3161,3162],{},"0.78",[623,3164,3165,3168,3171,3174,3177],{},[641,3166,3167],{},"qwen3.6-35b",[641,3169,3170],{},"0.80",[641,3172,3173],{},"0.73",[641,3175,3176],{},"0.84",[641,3178,3179],{},"0.65",[623,3181,3182,3185,3188,3191,3193],{},[641,3183,3184],{},"mistral-small-3.2",[641,3186,3187],{},"0.85",[641,3189,3190],{},"0.82",[641,3192,3157],{},[641,3194,3195],{},[65,3196,3197],{},"0.31",[58,3199,3200,3201,3204],{},"One model — Gemma — brackets its own English score in all three languages. One model fails Hausa outright. And the most instructive failure wasn't in the scores at all: ",[65,3202,3203],{},"mistral-small answered Bangla questions in English 32% of the time",", despite explicit instructions. Its content was fine; its language compliance was not. An LLM judge missed this consistently — a five-line script checking the answer's Unicode script caught it every time. Cheap deterministic checks beat model judges wherever you can write one.",[54,3206,3208],{"id":3207},"finding-2-retrieval-the-mirror-image","Finding 2 — Retrieval: the mirror image",[58,3210,3211,3212,3215],{},"Then the same questions went against the live corpus as ",[373,3213,3214],{},"queries",", testing whether our embedding model (bge-m3) can retrieve English passages from non-English text. The English control reproduced our production baseline exactly — always validate the harness before believing it.",[617,3217,3218,3231],{},[620,3219,3220],{},[623,3221,3222,3225,3228],{},[626,3223,3224],{},"Query language",[626,3226,3227],{},"MRR",[626,3229,3230],{},"Questions never retrieved",[636,3232,3233,3242,3250,3260],{},[623,3234,3235,3237,3239],{},[641,3236,3045],{},[641,3238,3157],{},[641,3240,3241],{},"0 of 40",[623,3243,3244,3246,3248],{},[641,3245,3133],{},[641,3247,3149],{},[641,3249,3241],{},[623,3251,3252,3254,3257],{},[641,3253,3136],{},[641,3255,3256],{},"0.66",[641,3258,3259],{},"2 of 40",[623,3261,3262,3264,3269],{},[641,3263,3139],{},[641,3265,3266],{},[65,3267,3268],{},"0.37",[641,3270,3271],{},[65,3272,3273],{},"15 of 40",[58,3275,3276],{},"Bangla retrieves at English parity. Hausa collapses — and machine-translation noise was ruled out by re-scoring on only the verified-clean translations (same result). The embedding model simply hasn't seen enough Hausa, which matches what the African-language retrieval literature (AfriMTEB) reports for this model class.",[58,3278,3279],{},"Put the two findings side by side and the architecture designs itself:",[300,3281,3284],{"icon":3282,"title":3283},"i-ph-arrows-left-right-duotone","The pivot",[58,3285,3286,3289,3290,3293,3294,3297],{},[65,3287,3288],{},"Hausa generates fine and retrieves badly — the exact opposite of our assumption going in."," So Hausa needs help only where the query meets the embedder. Translate the ",[373,3291,3292],{},"query"," for retrieval; answer ",[373,3295,3296],{},"natively"," from the evidence. The answer path never touches machine translation.",[58,3299,3300,3301,3304],{},"We measured that pivot the same afternoon: translating Hausa queries to English with NLLB-200 (the 1.3B distilled model, on CPU, seconds per query) lifted retrieval from ",[65,3302,3303],{},"0.37 MRR to 0.70 — above native Swahili"," — and cut never-retrieved questions from 15 to 2. A language we'd have called unservable in the morning had a proven end-to-end architecture by the evening, and the fix costs a CPU-seconds translation step on one language's queries.",[54,3306,3308],{"id":3307},"finding-3-redaction-the-traps-are-all-silent","Finding 3 — Redaction: the traps are all silent",[58,3310,3311],{},"The privacy layer was the biggest gap: the industry-standard PII framework has no models at all for these languages. We added a second analyzer service — transformer NER models trained on African-language and Bangla corpora — while leaving the existing, already-measured analyzer untouched. Two services, so that expanding coverage can never silently regress the thing you've already measured.",[58,3313,3314],{},"Measuring recall before deployment caught three defects, each of which would have been invisible in code review:",[2831,3316,3317,3327,3333],{},[493,3318,3319,3322,3323,3326],{},[65,3320,3321],{},"The redactor masked the word \"beneficiary\" and leaked the actual name."," The Bangla model's tokenizer strips vowel marks; under the default span aggregation the name fragments were dropped in alignment. One config value (",[2770,3324,3325],{},"aggregation: max",") fixed it — but only a recall measurement would ever have shown it.",[493,3328,3329,3332],{},[65,3330,3331],{},"The framework's context booster never fired."," ID patterns scored just below the redaction threshold while sitting next to a perfect context word (\"NID\", \"kitambulisho\"), because the multilingual tokenizer provides no lemmas for the booster to match. The numbers looked plausible; they were systematically one notch too low.",[493,3334,3335,3341],{},[65,3336,3337,3338,320],{},"\"Common Article 3\" became ",[2770,3339,3340],{},"\u003CPERSON_1>"," In Hausa, the NER model tagged the citation opener of a Geneva Conventions question as a person — which would have quietly destroyed one of the most important queries in the corpus. Domain-vocabulary allowlists exist for exactly this.",[58,3343,3344,3345,3348],{},"The deployed system now measures 120\u002F120 recall on our labelled corpus — and we publish that number with its caveat attached: the corpus is built from the same format rules the recognizers encode, so part of that score is tautological. When our Ukrainian\u002FRussian redaction met a genuinely held-out test set, ~100% became ",[65,3346,3347],{},"84.2%",". That's the number we quote for those languages, and the three new languages don't get a public number until their held-out sets exist. A recall figure without its provenance is marketing, not evidence.",[54,3350,3352],{"id":3351},"finding-4-measurement-pays-for-itself-in-bugs-you-werent-looking-for","Finding 4 — Measurement pays for itself in bugs you weren't looking for",[58,3354,3355,3356,3359],{},"Making the lexical search layer Unicode-aware (it previously dropped every Bengali character), we found the tokenizer had ",[65,3357,3358],{},"never matched the Ukrainian letter \"ї\""," — a one-character gap in a regex that had been silently fragmenting most Ukrainian words in keyword search since the feature shipped. Nobody noticed, because hybrid retrieval degraded gracefully instead of failing. That bug was found not by a bug report but by a probe refusing to produce a sensible number.",[58,3361,3362,3363,3366],{},"The same pattern recurred all day. Back-translation QA flagged 32 of 40 Hausa translations as broken — until we checked the ",[373,3364,3365],{},"checker"," and found the back-translation model was the one misreading Hausa (\"toilet\" became \"distance from home\"). Swap the reader, and 22 of 40 survive. Every measurement instrument is itself a thing to measure.",[54,3368,3370],{"id":3369},"the-self-hosting-question-finally-with-a-number","The self-hosting question, finally with a number",[58,3372,3373,3374,3377],{},"The last experiment of the session addressed a different standing question: what does self-hosting the production model actually cost in quality? We served the ",[373,3375,3376],{},"same model"," both ways — the hosted API at full precision, and a 4-bit quantized copy on our own GPU — and ran the identical evaluation, three times per arm.",[617,3379,3380,3393],{},[620,3381,3382],{},[623,3383,3384,3387,3390],{},[626,3385,3386],{},"Arm",[626,3388,3389],{},"Score",[626,3391,3392],{},"Run spread",[636,3394,3395,3406],{},[623,3396,3397,3400,3403],{},[641,3398,3399],{},"Hosted API (bf16)",[641,3401,3402],{},"0.872",[641,3404,3405],{},"0.008",[623,3407,3408,3411,3414],{},[641,3409,3410],{},"Self-hosted (int4)",[641,3412,3413],{},"0.809",[641,3415,3416],{},"0.038",[58,3418,3419,3422],{},[65,3420,3421],{},"Quantization to int4 costs about 6 points, outside the noise of both arms."," That converts a fuzzy architectural debate into a decision rule: don't route quality-sensitive traffic to an int4 arm at price parity. The self-hosted route is sold on what it's actually for — a sovereignty requirement a client pays for — with its measured quality cost stated up front, and FP8 on production-class hardware queued as the parity test.",[58,3424,3425,3426,3429,3430,3433,3434,3437],{},"One more instrumentation lesson: the first run of this comparison reported the delta as ",[373,3427,3428],{},"inside"," the noise floor. Twelve answers had scored zero because the ",[65,3431,3432],{},"judge's"," API rate-limited mid-run — an infrastructure failure masquerading as a quality signal. Only per-record failure logging made it visible. Re-judge those twelve, and the verdict flips. If your eval can't tell you ",[373,3435,3436],{},"why"," a record scored zero, it will eventually lie to you.",[54,3439,3441],{"id":3440},"what-it-cost-and-what-we-still-dont-know","What it cost, and what we still don't know",[58,3443,3444,3445,3448],{},"The entire investigation — two probes across four languages and three models, a retrieval pivot experiment, a deployed and verified redaction expansion, and a quantified self-hosting decision — cost ",[65,3446,3447],{},"single-digit euros in API tokens and zero GPU-hours rented",". The expensive-looking question (\"do we need a €1,000\u002Fmonth GPU?\") was answered for the price of a coffee, and the answer was \"not yet, and here is the number that will tell us when.\"",[58,3450,3451,3452,3455],{},"Just as important is what the evidence does ",[373,3453,3454],{},"not"," yet support, because publishing that list is what makes the rest credible:",[490,3457,3458,3461,3464,3467,3470],{},[493,3459,3460],{},"The 120\u002F120 redaction recall is a wiring proof, not a production claim — held-out sets per language come first.",[493,3462,3463],{},"Probe scores are not production scores; gold evidence makes generation easier than the live retrieval loop.",[493,3465,3466],{},"Translations passed machine QA only; native-speaker review is pending for anything user-facing.",[493,3468,3469],{},"Detection is measured on our evaluation register, not on colloquial chat, code-switching, or romanised Bangla.",[493,3471,3472],{},"Every number above is a mean over repeated runs with its spread — and the Hausa generation numbers carry more uncertainty than the others, because even two judges are still just two judges.",[58,3474,3475],{},"The method is the takeaway. Nothing here required rare infrastructure or a research budget. It is based on architectural decisions made from measurements — and treating every surprising number as a question about the instrument before believing it as a fact about the world.",[58,3477,3478],{},[373,3479,3480],{},"The full evidence pack behind this article — claim ledger, raw results, and the list of what each claim does and doesn't demonstrate — is part of our sovereign agent reference architecture.",[58,3482,3483],{},"As a bonus, we turned the same measure-first pipeline on the site itself: baena.ai is now machine-translated into Bangla, Swahili and Hausa — the same three languages this article is about — so this piece, and the rest of the site, are easier to reach for the people they're actually for.",[560,3485],{"description":3486,"icon":563,"link-label":564,"link-url":565,"title":2967},"If you're building AI for humanitarian or public-sector use where data residency and measurable safeguards are contractual, this is the working method we bring.",{"title":248,"searchDepth":249,"depth":249,"links":3488},[3489,3490,3491,3492,3493,3494,3495,3496],{"id":3024,"depth":249,"text":3025},{"id":3067,"depth":249,"text":3068},{"id":3111,"depth":249,"text":3112},{"id":3207,"depth":249,"text":3208},{"id":3307,"depth":249,"text":3308},{"id":3351,"depth":249,"text":3352},{"id":3369,"depth":249,"text":3370},{"id":3440,"depth":249,"text":3441},"2026-08-21T00:00:00.000Z","Three Languages, One Measured Pipeline — and what it taught us about measuring before building.",{"src":3500,"alt":3501},"https:\u002F\u002Fqecdwuwkxgwkpopmdewl.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fbaena-ai-assets\u002FArticles\u002FComfyUI_00030_.png","Illustration of three people browsing shelves in a library",{},"\u002Farticles\u002Fmultilingual-humanitarian-rag",{"title":3001,"description":3498},"3.articles\u002F9.multilingual-humanitarian-rag","Zt10wLKaUlVB7gEJdjakLvdmxapX6x2HbDIRNAnPkVA"]