who

I’m Teïlo Millet — a self-taught AI engineer based in Paris. I’m interested in research around large language models: reinforcement learning, alignment, safety, and everything in between.

now

consultant data & AI at OCTO Technology.

before

I tried to build enzu as a company at STATION F (Fighters Program). enzu is a toolkit for running LLM tasks with hard spending limits — budgets as laws of physics, not best-effort throttling. typed outcomes, async delegation, stress testing, and retry tracking. the idea: if you want to delegate a task to an LLM autonomously, you need guaranteed cost caps.

open source

textpolicy — reinforcement learning for text generation on Apple Silicon (MLX). GRPO, GSPO, LoRA fine-tuning.

enzu — budget-controlled recursive language model (RLM) execution in Python. hard caps on tokens, time, and cost. typed outcomes for every run.

gollm — unified Go interface for LLM providers. prompt optimization, structured output, model comparison.

raggo — production-ready RAG library in Go. document loading, semantic chunking, vector storage.

models

MiniMerlin-3B — reached global top 5 in the 3B parameter category on HuggingFace’s Open LLM Leaderboard.

writing

intuitus — a French newsletter on AI, prompt engineering, and the tech industry.

github · huggingface · x · linkedin · teilomillet@gmail.com