AI assistant · RAG
Freelance AI engineer · Forward Deployed EngineerAn assistant that cites its sources
An assistant that answers from your internal documents (procedures, contracts, technical documentation), cites the exact passage and says “I don’t know” rather than making things up. Each person’s access rights are respected, and it can be hosted on your servers, in your cloud or through an API.
What you get
The outcome
- Assistant wired to your documents, with each person’s access rights
- Measured evaluation on your real questions
- Hosted on your servers, in your cloud or via an API
- Timing
- 3 to 4 weeks
- You provide
- The documents involved and who can access them · Around twenty real questions your teams ask
- Price
- Quoted after the brief
What it is for
Three uses
The tasks that come up most often when a team gets in touch.
- 01
Technical documentation
Your teams find the right procedure or specification in one question, with a link to the source.
- 02
Contracts and legal
Find a clause or compare two versions, with the passage cited so you can check.
- 03
Internal support
The questions new starters keep asking, answered from your up-to-date documents.
How I work
Four steps
One contact, who does the work himself. You sign off before each next step.
- 01
Documents and rights
Which documents, who can access them, and around twenty real questions your teams ask.
- 02
Wiring it in
Your documents indexed with each person’s rights, in the environment of your choice.
- 03
Evaluation
The assistant is measured on your real questions: correct answers, cited sources, refusals when the information is missing.
- 04
Going live
Opened to your teams, with documentation and answer-quality monitoring.
Examples
See the work
Architecture, tests and measurements detailed on each page.
- AI engineeringVigieSovereign, auditable RAG for confidential documentsRun as a client mission, fictional client.
AI engineeringGrounded MemoryLine-level grounded RAG, MCP server, injections quarantinedStudio project, detailed on its page.
AI engineeringThe OracleShared data platform and forecasts evaluated over timeStudio project, detailed on its page.
FAQ
The questions
Not the answer? Write to me, I reply myself.
What is RAG?
Retrieval-augmented generation: before answering, the assistant searches your documents for the relevant passages and relies on them. It answers from your sources, not from memory, and can cite them.
How do you stop it making things up?
It only answers from the passages it retrieved and cites each one; it is evaluated on your real questions before going live, and when the information is missing, it says so.
Do our documents stay confidential?
Yes. The assistant can run on your servers, offline if needed, as I did at Thales Alenia Space; each person only sees the documents they already have access to.
How long does it take?
Three to four weeks, from document access to going live.