Freelancer profile translated to English.

Description

Machine Learning and Computer Vision Engineer, I design and deliver models that actually run in production, not prototypes.

For a multinational art logistics company, I single-handedly built a system that estimates the real dimensions (in meters) and 3D pose of an object from a single photo. Initially, there was no data: I sourced everything and built the dataset, then delivered three generations of models, reducing the average error from 40% to 17%.

I also developed, solo and end-to-end, a GenAI platform for an energy renovation player: LLM workflow (PydanticAI, Qwen), RAG and vector search (FAISS), MCP server, and a VAE trained on 13.8 million energy performance certificates (DPE), with anti-hallucination safeguards.

On each mission, I take care of the entire cycle: data research and preparation, training, evaluation, deployment (FastAPI, Docker, GPU, AWS/Scaleway).

Give me a problem: I find the data, I build the model, and I put it into production.

Languages

English
Native or bilingual
Spanish
Fluent
French
Native or bilingual

Workplace preferences

Remote only

Primarily works remotely

The Packengers
Machine Learning / Computer Vision Engineer (3D)
DIGITAL AND IT
January 2025 - Today (1 year and 6 months)
Paris, France
ML and Computer Vision Engineer for a multinational art logistics company. System that estimates the real dimensions (in meters) and 3D pose of an object from a single photo, to automate package measurement.
Initially, no data: complete data sourcing and building.
Three generations of models delivered, average error reduced from 40% to 17%.
Stack: diffusion (DDIM), foundation models (DINOv3, SAM 2), depth (Depth Anything 3), multimodal attention, uncertainty (NLL), Optuna.
Full cycle: data, training, evaluation, deployment (FastAPI, S3, H100 GPU, Docker, CI/CD, React dashboard).
Pytorch LLM Machine learning Deep Learning Computer Vision
Client confidentiel, rénovation énergétique
GenAI / LLM Engineer
DIGITAL AND IT
February 2026 - June 2026 (4 months)
Lyon, France
Design and development, solo, of a GenAI platform for an energy renovation player.
LLM workflow (PydanticAI, Qwen via Ollama/vLLM) generating renovation scenarios, with anti-hallucination safeguards (the LLM never produces figures).
VAE trained on 13.8 million DPEs for anomaly detection and latent signatures.
RAG search via FAISS on 13.8 million records for anchored recommendations.
MCP server exposing a dozen deterministic tools (regulatory energy calculation, aid, optimization).
Full-stack delivery: Python/FastAPI, Java/Spring Boot, Next.js/TypeScript, Terraform/Kubernetes on Scaleway.
Deep Learning Pytorch RAG LLM MLOps
Eden Games
Machine Learning Engineer and Cloud Developer
DIGITAL AND IT
September 2024 - February 2025 (5 months)
Lyon, France
AAA video game studio (Lyon).
Reinforcement Learning (RL) strategy for autonomous driving agents in racing games: training approach and feasibility study for integration into the game engine.
Setup of a cloud infrastructure in Infrastructure as Code (AWS CloudFormation, ECS, S3, IAM), containerized with Docker and CI/CD GitHub Actions.
Deployment foundations that I reuse today for ML model production.
AWS Fargate Pytorch MLOps Machine learning Deep Learning