I bridge the gap between AI research and production systems. No POCs left in a drawer. Real pipelines, real constraints, real usage.
I'm most at home where AI/ML meets backend engineering — where models have to actually run in production, under load, reliably.
- AI-powered backends — FastAPI services wiring up LLMs, agents and async task queues (Celery / Redis)
- LLM integration layers — RAG architectures, prompt pipelines, structured outputs, tool-calling loops
- End-to-end automation pipelines — ingestion → processing → storage → LLM → API, built to be reused
- Containerized, observable systems — Docker, GitHub Actions, self-hosted & cloud-ready deployments
A self-hosted cycling coach with a local LLM that never makes up a number — training-load
analysis, overtraining detection, and grounded AI advice.
FastAPI · React PWA · Ollama · SQLite · agentic tool-calling · self-hosted on a Raspberry Pi
🧠 Copain
A self-hosted, French-speaking assistant designed to absorb mental load, not pile more on —
voice, chat and geofence, semantic memory, pull-only by design.
FastAPI · Ollama · ChromaDB · PWA · mypy strict · 690+ tests · privacy-first
An LLM-powered automation pipeline (RSS → processing → storage → LLM summaries → REST API).
A reusable, production-grade pattern for AI automation.
Python · LLM · REST API · automation
Daily drivers: Python · FastAPI · LLMs (Ollama, RAG, agents) · Docker · PostgreSQL · SQL · GitHub Actions Also in the toolbox: Flask · Celery · Redis · Scikit-Learn · Hugging Face · Vector DBs (ChromaDB) Currently leveling up: Kubernetes / CKAD · MLflow · Airflow · PyTorch · cloud MLOps (AWS, Vertex AI)
- Production-first — reliability over cleverness. Code that's typed, tested and observable.
- Reliability is a habit, not an afterthought. Before software, I spent over 10 years in high-stakes industrial environments where a mistake had real consequences — that's where my attention to detail and my respect for constraints come from.
- Clear with everyone — I translate between technical and non-technical stakeholders.
- Autonomous and self-directed, comfortable owning a feature end to end in fast-moving teams.
Working full-time as a Python Developer – AI Integration at a French SaaS company, and leveling up my MLOps profile (Kubernetes / CKAD).
Always happy to talk shop about local LLMs, AI integration, or production backend design — if you're working on a hard problem at the AI × backend intersection, reach out.