AI & Data Engineer · ENSIAS × Télécom Saint-Étienne
Building AI systems enterprises can actually deploy.
I design and ship enterprise RAG pipelines, LLM integrations, and agentic systems — built from the start to run air-gapped, on-prem, and RGPD-compliant.
View GitHub ↗
About Me
I'm an AI & Data Engineer specializing in enterprise AI systems that organizations can confidently deploy and maintain — not just prototype. My work is shaped by real production constraints: tight compliance requirements, existing infrastructure, and teams that need to own the result long-term.
My core focus is enterprise RAG — hybrid retrieval architectures combining BM25, dense embeddings, and reciprocal rank fusion on pgvector, Qdrant, or FAISS — deployed on Kubernetes with full audit trails. I wire LLMs into existing Spring Boot and FastAPI backends via streaming microservice brokers, and build stateful agentic pipelines with LangGraph for complex workflow automation. Evaluation is baked in from day one: RAGAS, TruLens, and custom LLM-as-judge suites run on every deployment so quality regressions never reach users silently.
What sets my engagements apart is a focus on constrained environments: air-gapped clusters, Keycloak SSO, vLLM model serving, and zero external data egress — fully RGPD-compliant. I work fluently in French, English, and Arabic. If you're evaluating production AI and need an engineer who has already solved the hard parts — compliance, integration, evaluation — I take on a small number of engagements at a time. Let's talk.
enterprise AI
AI projects
FR · EN · AR
Hackathon 2022
What I Do
-
01
Enterprise RAG Systems
Hybrid BM25 + semantic + RRF pipelines on pgvector, Qdrant, or FAISS — on Kubernetes with auth and audit logging.
-
02
LLM Integration into Existing Backends
LLMs wired into Spring Boot, FastAPI, or Django — streaming, structured output, vLLM — inside your existing auth chain. No rewrite.
-
03
Agentic & Multi-Agent Systems
Stateful LangGraph agents, A2A coordination, and MCP integration for live tool access without context bloat.
-
04
LLM Evaluation & Quality Engineering
RAGAS, TruLens, and custom LLM-as-judge pipelines. Regression suites on every deployment so quality drops are caught before users see them.
-
05
Automated Data Ingestion & ETL
Web-scale document harvesting with delta detection. Structured knowledge bases ready for RAG indexing — no manual intervention.
-
06
Constrained & On-Prem Deployment
Air-gapped Kubernetes, Keycloak SSO, vLLM model serving — RGPD-compliant with zero external data egress.
-
07
Multilingual AI Systems (FR / EN / AR)
Cross-lingual RAG with language detection and multilingual-e5-large embeddings. I build and communicate natively in French, English, and Arabic.
-
08
PoC to Production
From validated prototype to hardened production system — with CI/CD, monitoring, rollback strategies, and documentation a team can own long-term.