Valentin Bernillon

I build and operate Python backends end to end. For the past months I have been working independently on a full-stack geospatial application: around 15 FastAPI REST endpoints with a modular service layer, Pydantic validation, structured logging and explicit error codes; a PostgreSQL/PostGIS data layer of 14 relational and spatial tables with spatial queries and SQL routines written and tuned by hand; ingestion, transformation and enrichment pipelines feeding it; and the deployment itself, on Docker and a Linux VPS behind a reverse proxy. I wrote the test coverage at every level — unit tests per module, SQL validation per table, automated end-to-end programs against the running stack — and I validate pipeline output rather than assuming a run that completes is a run that worked.

Part of that pipeline is LLM-assisted: model output is schema-validated and consistency-checked before it ever reaches the database. That is the part of AI integration I find genuinely interesting. The hard part is not calling the model, it is deciding what to do when it returns something wrong.

Before that, I spent six months as the sole engineer on my bachelor thesis at HEPIA: a real-time video platform on ARM64 embedded Linux, with no reference implementation to build on. I built the cross-compilation toolchain from scratch after finding the vendor framework had no support for it, deployed a CNN end to end from TensorFlow/Keras through ONNX and 8-bit quantisation to a runtime holding a 25 FPS / sub-40 ms budget, and diagnosed a broken vendor framework to root cause before evaluating alternatives. That work is where the habit of validating incrementally, with a checkpoint at every stage, came from — and I apply it directly to APIs and databases.

I learned to program without AI assistance and now use AI tools daily, reviewing generated code section by section and rewriting or rejecting it when it is wrong. I am most useful on problems that cross layers, where the path is not mapped out and the documentation is missing or wrong. I work autonomously, pick up new stacks quickly, and work in French, English and Spanish.

I am looking for a remote backend, data or AI-integration role on a team that writes tests and reviews code. Python, FastAPI, PostgreSQL/PostGIS, Docker, Linux, C/C++, Git.