Frαnçois fpaupier

Building AlexandrIA · Fractional Head of AI

LLM products · agents · RAG · hybrid retrieval · vLLM · LiteLLM · Langfuse · EU-sovereign cloud

I ship AI products that people actually use. I founded and operate AlexandrIA, a research co-pilot that turns messy questions into visual knowledge maps and client-ready deliverables. On the side I take a very small number of Fractional Head of AI mandates with post-seed startups that need someone to own the AI vision and make it real in production.

AlexandrIA

Live product: tryalexandria.fr · status

Knowledge workers use it to frame a problem, curate sources, and package insights. It is in production, with billing, EU-sovereign hosting, and recurring paying users. Built and operated end-to-end: multi-agent orchestration, RAG, LLM-as-judge evals, product, and infra.

Fractional Head of AI

For post-seed teams that already have a product and need an AI lead who has shipped, not a slide deck. I help set the roadmap, choose the stack, and stay close to delivery. Capacity is intentionally limited.

What I actually work in

Production LLM products, agents, RAG, knowledge graphs, and the infra that keeps them up.

Product / AI: TypeScript, Next.js, Python, FastAPI, multi-agent orchestration, tool-calling, RAG, hybrid retrieval, LiteLLM, vLLM
Eval & observe: LLM-as-judge, Langfuse, OpenTelemetry
Data / backend: Postgres, Supabase, embeddings
Ship & operate: Docker, Kubernetes, Ansible, GitHub Actions, EU cloud (OVHcloud, Scaleway), Cloudflare
Also in the field: on-prem / air-gapped copilots (self-hosted vLLM, Keycloak), knowledge graphs

I do not train computer-vision models in PyTorch day to day. I design, ship, and run LLM systems.

Background

Centrale Nantes, MSc Signal Processing & Statistics (top 10%). Engineering manager and ML engineer at the French DoD, then Schneider Electric and Kpler. Now founder of AlexandrIA, and AI lead on client production systems.

Guest Lecturer, LLM Pretraining & Post-training, SCAI (Sorbonne Université) & Centrale Nantes. CKA (Certified Kubernetes Administrator).

Teaching & speaking

Selected talks: TensorFlow World (Santa Clara, 2019) · AGIR 2024 for the Gendarmerie Nationale (LLM industrialization panel) · GenAI delivery at Qonto HQ (2025) · LLM sovereignty & supply chains, Centrale Nantes (2025). Catalog: fpaupier.fr/speaking

Post Mortem

I created, produced, and hosted Post Mortem for four years as producer and interviewer. Engineers walk through real incidents: outages, cyber, ML in production. 24 episodes, 2020–2024. On hold while I focus on shipping.