ReguLens

Client Snapshot

• Industry: International Policy / Public Sector
• ‍Client: United Nations — International Labour Organization (ILO)
• ‍Headquarters: Geneva, Switzerland
• ‍Timeline: 2025–2026
• ‍Team: 1 AI Architect · 2 Backend · 1 Frontend · 1 UX/UI · 1 PM
• ‍Services: AI Solutions · Platform Development · UX/UI Design
• ‍Stack: Claude (Anthropic) · Anthropic RAG · Claude Code · LangChain · pgvector · Pinecone · FastAPI · React · AWS Lambda · S3
• ‍Traction: Policy analysis reduced from days to 60 seconds

The Challenge

The ILO manages a sprawling corpus of international conventions, national labour laws, bilateral agreements, and policy frameworks — updated constantly across dozens of jurisdictions and languages. Policy analysts and advocacy teams were spending days, sometimes weeks, doing manual research before they could synthesize a coherent regulatory picture or draft an advocacy strategy.

The core problem wasn't access to information. It was the cost of making sense of it at scale. Every analysis had to start from scratch, and institutional knowledge walked out the door whenever a researcher moved on. The ILO needed a platform that could encode that expertise and put it to work in seconds.

• Regulatory volume: Thousands of conventions, national laws, and policy documents across 190+ countries, updated continuously
• Multilingual complexity: Source material spanning English, French, Spanish, Arabic, and more — with no unified query layer
• Advocacy bottleneck: Generating a defensible, evidence-backed advocacy strategy required days of expert synthesis before AI-assisted tooling existed
• Institutional knowledge risk: Research methodology lived in individual analysts' heads — not in a system that could scale
• Stakeholder trust: Outputs needed to be traceable, cited, and accurate enough to use in official UN-level advocacy contexts

‍

What Axented Did (Approach & Delivery)

We treated ReguLens as a knowledge infrastructure problem, not just a chat interface. The goal was to build a system that could ingest the full breadth of ILO regulatory content, reason over it with high accuracy, and surface analysis that analysts could actually use and cite.

AI-First Architecture

• LLM-native pipeline design: Designed the core reasoning layer around large language models with structured prompting, retrieval-augmented generation (RAG), and citation enforcement — so every output is traceable to a source document
• Regulatory knowledge graph: Ingested, chunked, and embedded thousands of ILO conventions, national law summaries, and policy briefs into a vector store optimized for fast, high-precision retrieval across jurisdictions
• 60-second analysis engine: Built the query-to-output pipeline to compress multi-day manual research into under a minute — including jurisdiction scoping, cross-reference resolution, and conflict detection
• Multilingual retrieval: Implemented language-agnostic document indexing so analysts can query in their working language and retrieve relevant content across language boundaries
• Advocacy strategy generation: Developed a structured output module that generates defensible, evidence-backed advocacy strategy drafts aligned with ILO policy objectives and stakeholder audiences

Platform Development

• Backend architecture: Built a robust REST API layer managing document ingestion, embedding, retrieval orchestration, and session state — designed for reliability at UN-scale usage
• Document processing pipeline: Engineered an automated ingestion flow that handles new conventions and law updates without manual intervention, keeping the knowledge base current
• Audit trail & citation system: Every AI-generated output links back to specific source documents with structured citations — meeting the traceability requirements of an international policy environment

UX/UI Design

• Analyst-first interface: Designed the query and results experience around the actual workflow of a policy analyst — not around the underlying AI architecture
• Jurisdiction scoping UI: Built intuitive controls for narrowing analysis by region, industry, and convention type — so outputs are specific and actionable, not generic
• Export and sharing flows: Designed output formats suitable for internal reports, advocacy briefs, and stakeholder presentations

‍

Solutions Implemented

AI Core

RAG-powered regulatory engine

Retrieval-augmented generation pipeline that queries a purpose-built vector store of ILO conventions and national laws — returning cited, jurisdiction-specific analysis in under 60 seconds.
‍

Advocacy strategy generator

Structured LLM module that synthesizes regulatory context into draft advocacy strategies, aligned with ILO policy objectives and tailored to specific stakeholder audiences.

Multilingual retrieval layer

Language-agnostic document indexing and query resolution — enabling analysts to work across English, French, Spanish, and Arabic source documents without manual translation overhead.

______
‍

Infrastructure

Automated document ingestion pipeline

Processing pipeline that ingests, chunks, embeds, and indexes new regulatory content automatically — keeping the knowledge base current as ILO standards and national laws evolve.

‍

Citation and audit trail system

Every AI output is linked to specific source documents with structured citations — meeting the traceability and accountability requirements of an official UN policy environment.
______
‍

UX/UI

Analyst-first query interface

Clean, purpose-built UI designed around the real workflow of a policy analyst — jurisdiction scoping, context-setting, output review, and export — not around the AI architecture underneath.

‍

Results

ReguLens launched as a flagship initiative inside the ILO and was publicly announced by the organization as a new standard for how AI can be applied to international labour policy work. The results speak to what happens when AI infrastructure is built thoughtfully for a high-stakes institutional environment.
‍

60s Policy analysis that previously required days of expert research, now delivered in under 60 seconds

190+ Member state jurisdictions covered by the regulatory knowledge base

100% AI outputs traceable to cited source documents — meeting UN-level traceability requirements
‍

• Institutional adoption: Publicly announced by the ILO as a production tool for policy analysis and advocacy strategy generation at the UN level
• Workflow transformation: Compressed multi-day research cycles into a single 60-second query — freeing analysts to focus on strategy rather than retrieval
• Scalable knowledge base: Regulatory content is indexed and updated automatically — institutional knowledge no longer depends on individual researchers
• Multilingual reach: Analysts working across ILO's official languages can access and query content without translation bottlenecks
• Defensible outputs: Citation-enforced AI generation means outputs are usable in official policy and advocacy contexts from day one

‍

Team & Delivery Model

AI Architect / Tech Lead

LLM pipeline design, RAG architecture, embedding strategy, retrieval optimization, and overall AI system integrity.

Backend Engineers (×2)

API layer, document ingestion pipeline, vector store management, session orchestration, and performance hardening.

Frontend Engineer

React application, query interface, output rendering, citation display, and export flow implementation.

UX/UI Designer

Analyst workflow research, information architecture, interface design, jurisdiction scoping controls, and usability testing.

Project Manager

Sprint planning, ILO stakeholder coordination, milestone governance, and cross-functional delivery alignment.

Operating Model

Agile, milestone-driven delivery with iterative releases, continuous prioritization, and tight stakeholder feedback loops across Geneva and LATAM time zones.

‍