Alexandre Martel

I design and deploy generative AI systems in production, on your infrastructure, in open source, without dependence on a proprietary provider.

Many providers plug in an API and call it AI. I do the opposite: I deploy your own models (Llama, Qwen, Mistral), serve them with vLLM/SGLang on Kubernetes, and cut your inference costs by 2 to 3 thanks to real GPU expertise (CUDA, Triton, quantization). Your data stays with you, GDPR compliant.

What I deliver:

> Deployment of open-source LLMs in production (vLLM, SGLang, Kubernetes)

> GPU inference cost optimization (quantified benchmark before/after)

> Robust RAG pipelines (LlamaIndex, Qdrant, sourced answers)

> Multi-step agents in production (LangGraph, MCP)

> Fine-tuning of specialized models

> Migration from proprietary API to self-hosted

Background: Centrale Lille, Applied Mathematics Master's, AI research at Tohoku University (Japan). AI Engineer at Equans/Bouygues (multi-cloud agents, computer vision, fine-tuning). Formerly at Capgemini Invent (AI for major financial players). Winner of the PyTorch x GPU MODE Hackathon 2026 (LLM optimization on B300 cluster).

Describe your needs, and I'll tell you frankly if I'm the right person.