Data & AI Apprentice @ SNCF • ECE Paris (Ingénieur cycle)
I design pragmatic AI systems, from retrieval-augmented pipelines to autonomous agents that can speak SQL. Previously at Tetra7, I implemented a lightweight Model Context Protocol (MCP) setup: an MCP server exposing tools, resources, and prompts to the model client. I paired it with RAG (embeddings + vector search), a schema-aware SQL agent for safe, read-only queries, and a production-grade document classifier. I’m currently an apprentice at SNCF Voyageurs and I’m seeking a 3-month internship (May–July 2026).
Built an MCP server that advertises tools, resources, and prompts to an MCP client with capabilities. Implemented tool discovery. Integrated a RAG layer (embeddings + vector index) to serve contextual resources, enabling the agent to select and invoke the right tool per request.
Built clean Laravel backends with RESTful endpoints, request validation, policies, and expressive Eloquent relations. Shipped Livewire dashboards with server-driven interactions, eager-loaded metrics, and background queues (Redis) for long-running jobs. Added caching and structured logs to keep p95 fast and debugging predictable.
Production classifier for industry documents with robust PDF/text parsing, field normalization, and calibrated confidence scores. Added an evaluation harness (stratified sets, error breakdowns) and simple thresholding to balance precision/recall. Exposed the model behind an HTTP endpoint with idempotent requests and audit trails for reliable ops.
AMOA within the SI Essieux (DSI Matériel). Bridging business needs and data/AI features, writing specs, and aligning stakeholders around reliable, auditable solutions.
Laravel, PHP, Python, FastAPI, SQL (MySQL/PostgreSQL), ClickHouse basics, Redis, queues.
RAG pipelines, embeddings, vector search, prompt design, evaluation, lightweight agents.
Docker, Plesk, Herd, GitHub Actions, monitoring and structured logging.
Paris or remote. I value clear specs, measurable outcomes, and fast feedback loops.