Research Engineer - Tokenizer & Protocols

ABOUT MISTRAL Mistral provides full-stack AI solutions: from frontier models to developer tools, applications, and compute. We partner with enterprises tackling the hardest problems across high-stakes industries like finance, manufacturing, defense, healthcare, and the public sector, co-creating customized AI systems that they can run on their terms. We are a dynamic, collaborative team passionate about AI and its potential to transform society. Our diverse workforce thrives in competitive environments and is committed to driving innovation. Our teams are distributed between Europe, North America, Asia and the Middle East. We are creative, low-ego and team-spirited. ABOUT THE ROLE Between a conversation and a model sit the formats and tokenizers that every part of our stack relies on. You will help own them, and the tooling around them, so that training, evaluation and production all speak exactly the same language. You will work across the entire stack, spanning pretraining, post-training and deployment, and your decisions will reach every model we ship. WHAT YOU WILL DO - Design, build and maintain core libraries for tokenization and model interfaces. - Evolve our formats and tokenizers as models and use cases change. - Turn research needs into robust, shared tooling. WHAT WE'RE LOOKING FOR - Strong software engineering background, with excellent Python. - Excellent understanding of tokenizers: BPE and other tokenization algorithms, special tokens, chat templates, and tokenizer training; ideally hands-on experience with libraries such as tiktoken or Hugging Face tokenizers. - Experience building production-grade libraries used by many people. - A taste for correctness and attention to detail. - Good knowledge of how LLMs are trained and served. - Clear communication, low ego, and a collaborative mindset. - Nice to have - Open-source maintenance experience. - Experience with inference engines. LOCATION AND REMOTE This ro