New AI Scientist , Paris/london, Onsite or Hybrid or Remote

Mistral AI Mistral AI · AI Frontier · Paris, France · Research

AI Scientist role at Mistral AI focused on researching and developing novel methods to advance LLM capabilities, including reasoning, code generation, and agent-based systems. The role involves training, evaluating, and analyzing AI models at scale, working with multimodal data, and optimizing distributed systems. Emphasis on contributing to open-source frontier models and shipping AI systems with tangible impact.

What you'd actually do

  1. Research and develop novel methods to advance the capabilities of LLMs.
  2. Design and implement tooling and infrastructure for training, evaluating, and analysing AI models at scale.
  3. Collaborate with scientists, engineers, and product teams to ship AI systems with tangible impact.
  4. Explore and innovate across use cases such as reasoning, code generation, and agent-based systems.
  5. Work with multimodal data, including text, image, and speech, to expand model capabilities.

Skills

Required

  • PyTorch or JAX
  • Ray or Kubernetes
  • Python, Rust, Go, or Java
  • design complex software systems
  • transition them into production environments
  • training large transformer models in a distributed setting
  • full MLOps stack, including fine-tuning, evaluation, and deployment
  • building scalable and reliable systems

Nice to have

  • contributing to relevant scientific domains, such as publications or open-source projects

What the JD emphasized

  • track record of contributing to relevant scientific domains, such as publications or open-source projects
  • Experience with training large transformer models in a distributed setting
  • Familiarity with the full MLOps stack, including fine-tuning, evaluation, and deployment

Other signals

  • developing and scaling cutting-edge models
  • democratizing frontier intelligence
  • open, high-impact AI systems
  • shipping AI systems with tangible impact
  • contribute to the open-source community by releasing frontier models