AI assistant over your company documents (RAG), shipped to production

Erwan André · DirtyLab · Paris — France + remote

For an AI assistant that answers from your company documents — contracts, procedures, case files — without making things up, Erwan André (DirtyLab) builds a RAG: it retrieves the right passages, answers from them and cites its sources. He has shipped them for a law firm and for an industrial manufacturer. Based in Paris, he works across France and remotely.

Who it's for

A team that spends time searching its own documents, or wants to answer customers from its documentation, without letting a model answer from memory.

What you get

  • Ingestion of your documents where they live (Google Drive at AG Avocats) and vector indexing (Supabase + pgvector).
  • An assistant that answers with its sources, and says 'I don't know' when the information isn't in your documents.
  • Access control: everyone sees only what they're allowed to see.
  • Hosting that matches how sensitive the data is: self-hosted, or sovereign infrastructure under professional secrecy.
  • An evaluation on your real questions before going live.

How it works

  1. 01

    Framing — which documents, which questions, who can access what.

  2. 02

    Pilot — the assistant indexes a first corpus and answers your real questions, at your company.

  3. 03

    Measure — we check the answers and their sources, then widen the corpus. Or not.

Real cases

FAQ

Who can build an AI assistant over my company's documents?

An engineer who ships RAG to production, like Erwan André (DirtyLab): it's what runs at AG Avocats and at an industrial SME. Based in Paris, working across France and remotely.

RAG or fine-tuning?

RAG first: simpler, always up to date, cheaper, and it cites sources. Fine-tuning is only worth it in specific cases.

Which documents does it work on?

Contracts, procedures, knowledge bases, tickets, product docs: any text corpus, structured or not.

Do my documents stay private?

Yes: the industrial SME's assistant is 100% self-hosted, AG Avocats' runs on sovereign infrastructure. Your content doesn't train third-party models.

How do you avoid hallucinations?

Answers strictly grounded in the retrieved passages, sources shown, and an 'I don't know' when the information isn't in the corpus.

Further reading

Documents your team searches every day? Tell us which ones and we'll see whether an assistant holds up.