RAG on your own documents
Connect the documents your team already uses. Preserve permissions, retrieve the right evidence and make every answer traceable to its source.
AI engineering
I'm Adam Boudjemaa, an AI engineer in France. I build production AI agents and retrieval-augmented generation (RAG) systems. Each one comes with evaluations, permissions and a handover your team can use.
Connect the documents your team already uses. Preserve permissions, retrieve the right evidence and make every answer traceable to its source.
Build bounded tools and approval steps around reconciliation, reporting or operations. Define what the agent can do and when a person takes over.
Test missing evidence, wrong numbers and tenant boundaries. Add abstention, cost limits and regression checks before giving a workflow more autonomy.
I delivered an 11-agent platform for a confidential US real-estate fund. The work included FastAPI, Postgres/pgvector, document ingestion, verified citations and numeric checks. Eleven describes the initial delivery.
Read the delivery case studyAs CTO at Integra, I built the engineering team and led infrastructure through launch preparation. Because of that work, I treat permissions, failures and handover as part of the design in production AI.
Review my experience and CVYes. I'm open to permanent AI engineering roles and scoped freelance missions. Send the role or business problem, your stack, your timeline and your budget or salary range. We then check whether my experience matches.
I'm based in France, an EU citizen, and work in French and English. I'm open to remote work, on-site engagements and relocation, depending on the role.
I delivered a production multi-agent platform for a confidential US real-estate fund. It included RAG on FastAPI and Postgres/pgvector, Amazon Bedrock, document extraction, verified citations, numeric checks and explicit abstention. The public case study describes the initial 11-agent delivery without naming the client.
Yes. I start by inspecting one workflow, its source data and its failure cases. Then we agree on a release gate. This review may lead to a retrieval fix, a smaller agent design or stronger evaluations before adding features.