SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Mistral is seeking an Applied Scientist to drive innovative AI research and collaborate with enterprise clients on complex, high-impact projects. You will develop state-of-the-art models across multiple modalities (text, image, speech) and apply them to diverse use cases in finance, manufacturing, defense, healthcare, and the public sector.
Key responsibilities include:
- Running pre-training, post-training, and deployment of SOTA models on large GPU clusters (thousands of GPUs), managing distributed training challenges like OOM errors and NCCL communication issues
- Generating, curating, and evaluating datasets for pre-training and post-training to ensure models exceed performance expectations
- Developing tools and frameworks for data generation, model training, evaluation, and deployment
- Collaborating cross-functionally with internal and external science, engineering, and product teams to build agents, RAG pipelines, and custom AI solutions
- Managing research projects and maintaining communication with client research teams
You are fluent in English and Korean with excellent communication skills, able to explain complex technical concepts to both technical and non-technical audiences. You are an expert in PyTorch or JAX, comfortable navigating large codebases independently, and write clean, high-performance, fault-tolerant Python code. You are self-directed, low-ego, collaborative, and have a proven track record of success through personal projects, professional work, or academia.
Ideal candidates hold a PhD or Master's in a relevant field (Mathematics, Physics, Machine Learning) or demonstrate exceptional capability from another background. Additional strengths include research experience in agents, multi-modality, robotics, diffusion, or time-series; contributions to large open-source or industry codebases; publications in top academic venues; and a passion for code quality (typing, testing, CI/CD).
