Research Engineer, Forge

About Mistral

Mistral delivers comprehensive AI solutions spanning advanced models, developer tools, and applications, catering to complex challenges in critical industries such as finance, manufacturing, defense, healthcare, and public sectors. Their collaborative, innovative global team spans Europe, North America, Asia, and the Middle East, maintaining a creative and low-ego culture.

Role Overview

As a Research Engineer within the Forge project, you will translate real-world client needs into robust training and deployment workflows, spanning model adaptation and post-training tasks (including CPT, SFT, RL, and distillation), evaluation, data handling, and infrastructure. This role merges experimental research with production realities and involves collaboration across science, engineering, product, and customer teams to deliver dependable, maintainable Forge solutions.

Key Responsibilities

• Develop and refine workflows for post-training and evaluation processes such as CPT, SFT, RL, and distillation, transforming prototypes into repeatable Forge "recipes".
• Build tools and pipeline systems for synthetic data generation, data management, training, evaluation, and deployment.
• Troubleshoot and strengthen large-scale machine learning systems, addressing distributed training challenges, scheduling executions, checkpointing, observability, and ensuring reproducibility.
• Enhance the Forge codebase by creating clear APIs, conducting tests, writing documentation, and designing maintainable abstractions.
• Advance the reinforcement learning training infrastructure, focusing on scalable asynchronous rollouts and high-throughput systems suitable for frontier-scale models.
• Ensure Forge deployment processes are smooth and flexible, adaptable to diverse client environments, including varied hardware, software stacks, cloud, and on-premises settings.
• Collaborate with researchers and infrastructure engineers to identify system bottlenecks and implement practical improvements.

Candidate Profile

• Proficient Python software engineering skills with experience managing extensive codebases, including testing, code review, continuous integration, and operational responsibility.
• Hands-on experience with machine learning frameworks like PyTorch or JAX.
• Strong foundation in systems and infrastructure principles.
• Practical knowledge of large language model (LLM) training or post-training techniques such as fine-tuning, reinforcement learning, distillation, evaluation, and managing data pipelines.
• Excellent problem-solving and debugging capabilities in complex and ambiguous systems, including distributed jobs, data inconsistencies, quality regressions, and infrastructure failures.
• Effective communicator able to engage with both technical and non-technical audiences.
• Highly proactive, humble, and adaptable when working in rapid, evolving, and loosely defined environments.

Preferred Qualifications

• Experience with distributed training frameworks such as FSDP, DeepSpeed, or Megatron.
• Familiarity with cluster management and orchestration tools like SLURM, Ray, Kubernetes, Kueue, Karpenter, or Skypilot.
• Background in creating reliable ML infrastructures, evaluation mechanisms, or handling large-scale data processing pipelines.
• Research experience related to LLMs, agent-based models, multimodal systems, reasoning, coding tasks, or domain adaptations.
• Contributions to open-source projects, scholarly publications, or development of widely adopted internal tools.
• Exposure to training models with billions of parameters at pre-training or reinforcement learning stages and working with extremely large datasets (petabyte to exabyte scale).
• Ability to analyze and resolve systemic bottlenecks from foundational principles.

Benefits

Mistral offers a comprehensive benefits package aimed at supporting employee well-being, career growth, and work-life balance. The range of benefits varies by country and may include healthcare, parental leave, retirement plans, relocation assistance, wellness initiatives, meal and transport allowances, among other location-specific incentives.

Privacy Notice

Candidate privacy is respected and protected; more details are available in Mistral's Applicant Privacy Policy.