@guillaume-flambard2
Freelance Full-Stack & AI Engineer, Memo Labs
1h ago · AI Engineering · Paris

Turning a hostile EU portal into data you can actually search

The EU Register of Nutrition and Health Claims has a web UI and no clean export. I built sku-regulator to see how far a small RAG pipeline could go on it.

What it does:

  1. Pulls the register from the same backend API the official UI uses (reverse-engineered).
  2. Flattens its recursive value tree into one clean record per claim.
  3. Stores claims, products and per-country status in Postgres + pgvector.
  4. Answers questions with hybrid search: full-text and embeddings fused with reciprocal rank fusion, every answer tied to its source record.
  5. Sends low-confidence extractions to a human review queue. Wrong data here has consequences, so people make the call.

The repo even records its own 70-second demo: Playwright drives the UI, Remotion adds titles, ffmpeg mixes the sound.

It's a prototype and an engineering case study, not legal advice.

Through Memo Labs I do freelance work on exactly this: messy data sources, AI features tied to real sources, n8n automations, web apps shipped end to end. If your team has a scraping, RAG or automation problem stuck in the backlog, send me a message.

What's the worst data source you've had to build on?

github.com/guillaume-flambard/sku-regulator

RAG

Comments

No comments yet
Start the conversation.
Log in to join the thread.