The International Labour Organization manages one of the most complex regulatory knowledge bases in existence: thousands of international conventions, national labour laws, bilateral agreements, and policy frameworks across 190+ member states, continuously updated, in multiple languages.
Before ReguLens, a policy analyst wanting to understand how a specific labour standard applied across ten jurisdictions would spend days — sometimes weeks — reading source documents, cross-referencing updates, and synthesizing findings before drafting an advocacy strategy.
In 2025, Axented partnered with the ILO to change that. The result was ReguLens: a production-grade AI platform that delivers cited, jurisdiction-specific regulatory analysis in under 60 seconds. It was publicly announced by the ILO as a new standard for how AI can be applied to international labour policy.
The ILO's challenge wasn't access to information — the documents existed. It was the cost of making sense of them at scale. Every analysis started from scratch. Institutional knowledge lived in individual researchers' heads and walked out the door when those researchers moved on.
What the ILO needed wasn't a better search engine. It needed a system that could encode regulatory expertise and put it to work on demand. That distinction — knowledge infrastructure versus search — shaped every architectural decision we made. A search interface returns documents. ReguLens reasons over them: it scopes jurisdictions, cross-references conventions, detects conflicts between national laws and international standards, and generates defensible, evidence-backed outputs that analysts can use in official UN advocacy contexts.
We built the AI layer around retrieval-augmented generation with strict citation enforcement. Every output from ReguLens links back to specific source documents. This wasn't a nice-to-have. It was a hard requirement: UN-level policy work requires traceability. An AI output that can't be verified against a source document isn't usable in an official advocacy context, regardless of how accurate it appears.
The vector store was built on pgvector and Pinecone, with an embedding pipeline designed for multilingual retrieval — analysts working in French can query content indexed in Spanish and English without manual translation overhead. The backend was built in FastAPI, deployed on AWS Lambda for reliability at variable UN-scale usage patterns.
The core user experience — query to cited output — runs in under 60 seconds. That pipeline involves: jurisdiction scoping (which of the 190+ member states are relevant to this query?), document retrieval (which conventions, national laws, and policy briefs apply?), cross-reference resolution (are there conflicts between what different jurisdictions require?), and output synthesis (what does a policy analyst need to know, in what format, to act on this?).
Building that pipeline to be both fast and accurate required careful work on chunking strategy, retrieval ranking, and the structured prompting layer that turns retrieved context into usable output. The prompting work alone went through more than a dozen iterations against a test set of real ILO analyst queries before the accuracy met the bar for production.
ReguLens launched as a flagship initiative inside the ILO with full institutional adoption: publicly announced, integrated into the policy analysis workflow, and positioned as the new standard for AI-assisted regulatory research. Policy analysis that previously required days of expert research now takes under a minute. The regulatory knowledge base updates automatically as new conventions and national laws are indexed. Outputs are 100% traceable to cited source documents.
The ILO analyst who described the result put it plainly:
"ReguLens turned our regulatory knowledge base into an active tool — one that gives our analysts the answers they need in seconds, with the citations to back them up. It changed how our teams work."
The platform was delivered by a six-person Axented team: one AI architect who owned the LLM pipeline, RAG architecture, and embedding strategy; two backend engineers for the API layer and document ingestion pipeline; one frontend engineer for the React application; one UX/UI designer for the analyst-facing interface; and a project manager handling ILO stakeholder coordination across Geneva and LATAM time zones.
The engagement ran through 2025–2026. The delivery model was agile with milestone-driven releases and tight feedback loops with ILO stakeholders — the kind of iteration that's required when the users have institutional knowledge the engineering team needs to encode correctly.
Your next big thing starts with a call. 30 minutes. We scope your idea, recommend the right service, and answer every question. No pitch deck, no pressure.
.svg){kind=icon}
%20(1)-2.svg)
