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.
The team
You will join the AI4Engineering Science team, the Mistral research group that builds the data and evaluation infrastructure and owns the SFT/RL processes for AI4Engineering that teach our frontier LLMs to do real engineering work. The agentic engineering group works alongside the Physics AI research in the wider Science organisation.
This seat sits in the agentic CAD stream. Together with the Solutions and Applied teams, the group carries this work into customer projects with industrial partners in automotive, aerospace, semiconductors, and beyond.
The Role
Mistral is looking for an AI Scientist with deep machine learning expertise and hands-on CAD experience to expand what our agentic tools can do for computer-aided design. Your focus is a single mission: teaching Mistral's LLMs to generate, analyze, and modify CAD designs the way an expert design engineer would.
Your core work is building the pre- and post-training data for Mistral's LLMs to reason about and execute real CAD tasks: text-to-CAD and image-to-CAD generation from technical drawings, geometric analysis and QA of CAD parts, and modification of existing designs. Because Mistral trains its own frontier LLMs, the data and verifiers you design ship directly into models you can hold - a rare position, and the core of the job.
Alongside this, you will help shape the agent architectures and harnesses that let these models operate reliably inside multi-step CAD workflows, driving CAD tools through APIs or computer use across the whole design loop.
This is early-stage work, and that's the point: you are joining at the foundation, shaping the data, verifiers, and agent scaffolding that decide whether these systems become reliable or stay demo-grade. There is no inherited playbook; you will help define what good looks like, and your work will set the direction the CAD stream builds on rather than extend an existing one.
What You Will Do
Design pretraining, SFT, and RL data for CAD tasks: text-to-CAD and image-to-CAD generation from technical drawings, geometric analysis and QA of CAD parts, and design modification
Define verifiers and evaluation criteria that capture what "correct" and "high-quality" actually mean for CAD tasks
Scale the data pipelines that produce agentic CAD training traces
Design and improve agent architectures and harnesses: how models plan, call CAD tools via API or computer use, recover from errors, and chain steps together across long-horizon design workflows
Build and maintain CAD evaluation benchmarks and diagnostic tooling, from parametric CAD and drawing-to-3D benchmarks to industrial evaluations, identifying where models fail and tracing those failures back to gaps in data, reward design, or agent scaffolding
Collaborate with CAD domain experts and the CAE and EDA streams (and the science and solutions teams more broadly) to identify which CAD workflows are highest-value to target next
Contribute to Mistral's broader pre- and post-training research, sharing findings and methodology across the science organization
What We're Looking For
Deep, hands-on machine learning expertise, particularly LLM development: you have built LLM training pipelines (SFT, RL, fine-tuning, GRPO), demonstrated in publications or shipped systems rather than CV claims
Demonstrated hands-on CAD experience: you have designed, scripted, or automated real CAD workflows in tools such as SolidWorks, CATIA, Siemens NX, FreeCAD, or CADQuery, ideally with CAD-kernel grounding (OpenCASCADE, build123d)
Fluent English with excellent communication skills: you can explain technical ML and CAD concepts to both engineering and non-technical audiences
You write clean, readable Python code and are comfortable in Linux/HPC environments
Self-directed: you do not need detailed roadmaps to make progress
Low-ego, collaborative, and eager to learn at the intersection of CAD and ML
It would be great if you:
Have experience building or fine-tuning agentic systems (tool use, multi-step planning, agent orchestration frameworks)
Have experience with reward modeling, RLHF/RLAIF/RLVR, or preference-based training
Have first-author publications or open-source benchmark work in agentic CAD or CAD-ML, such as CAD generation benchmarks or RL/GRPO fine-tuning of vision-language models for CAD
Have contributed to a large open-source or industry codebase
Have publications in engineering and ML venues (NeurIPS, ICLR, ICCV, SIGGRAPH, JFM, AIAA, etc.)
What We Offer
We offer a comprehensive benefits package designed to support your well-being, growth, and work-life balance. Benefits vary by country and may include healthcare coverage, parental leave, retirement plans, relocation support, wellness programs, meal and transportation allowances, and other location-specific perks.
For the most up-to-date details on benefits available in your location, please refer to our Benefits page.
Privacy Policy
Your privacy matters to us. You can learn more about how we handle your personal data in our Applicant Privacy Policy.