Software Engineer, Reinforcement Learning

Description

The company is building frontier models and systems that are used by millions of people through Copilot and the wider Microsoft ecosystem. The Reinforcement Learning Engineering team builds the environments, data, evaluations, training systems and algorithms that improve models after pre-training. Work covers reasoning, instruction following, tool use, coding and agentic knowledge work.

The team runs the full reinforcement learning loop: converting product and research goals into measurable tasks, gathering high-quality experience and feedback, training at scale, assessing model behavior and quickly turning results into the next model generation.

This is a hands-on engineering role that blends strong software work with applied machine learning. You will set technical direction and take ambitious research-to-production efforts from strategy to production impact. You will build dependable reinforcement learning systems that work at frontier-model scale. Day to day you will work with researchers, evaluation teams, infrastructure engineers and product partners to turn vague model and product goals into concrete capability gains. The role suits someone who moves comfortably between algorithms and systems, forms clear hypotheses and owns work end to end.

The ideal candidate cares about reinforcement learning and post-training for large language models, works well in a collaborative, quick-moving team with demanding goals, takes care over craft and detail, relies on rigorous experiments and measurement, owns work from start to finish, and wants to ship models and capabilities to users.

Responsibilities

• Set technical direction and lead multi-quarter efforts to improve large language models and agents with reinforcement learning.
• Build and scale RL environments, rollout systems, training pipelines and data generation workflows.
• Create high-quality tasks, feedback signals, reward models and training datasets from human input, model outputs and real-world interactions.
• Design hypotheses and run experiments to improve reasoning, instruction following, tool use, coding and other agentic abilities.
• Work with evaluation teams to define benchmarks, find causes of model failures and measure core and real-world performance.
• Make large-scale RL experiments more reliable, observable, fast and reproducible.
• Analyze experiment results, share findings and use them to guide model, data, evaluation and product roadmaps.
• Work with research, infrastructure, safety and product teams to turn new techniques into models used by millions of people.
• Mentor engineers, improve experimental and system design, and help build an inclusive engineering culture based on knowledge sharing and technical excellence.

Requirements

• Bachelor's degree in computer science or a related technical field and 6+ years of engineering experience coding in languages such as C, C++, Java, JavaScript or Python, or equivalent experience.
• Preferred: a Master's degree in computer science or a related field and 8+ years of engineering experience coding in languages such as C, C++, Java, JavaScript or Python, or a Bachelor's degree in computer science or a related field and 12+ years of such experience, or equivalent experience.
• Preferred: experience building reliable machine learning systems and applying reinforcement learning, reward modeling, preference optimization, post-training or closely related methods to large language models or other large generative models.
• Preferred: a track record of leading technically ambitious machine learning initiatives across multiple teams, from hypothesis and experimental design through measurable model or product impact.
• Preferred: experience shaping technical direction across organizational boundaries and mentoring engineers, using quantitative evidence to decide and deliver strong results in unclear situations.
• Preferred: experience building RL environments, agent harnesses, simulators or large-scale rollout infrastructure.
• Preferred: experience with distributed training or inference on accelerators such as GPUs.
• Preferred: experience generating, curating, filtering and versioning large training datasets.
• Preferred: experience building reward signals from human feedback, model feedback, verifiers or real-world outcomes.
• Preferred: experience evaluating agents on reasoning, coding, tool use or long-horizon tasks.
• Preferred: familiarity with modern post-training methods and open-source machine learning frameworks.
• Preferred: clear written and spoken communication, including explaining complex experimental results to cross-functional partners.
• Preferred: experience shaping an RL, post-training or evaluation roadmap across several teams or model releases.

Conditions

Software Engineer IC5 role.

Typical base pay across the U.S. is USD $142,800 to $274,800 per year. In the San Francisco Bay Area and New York City metropolitan area, base pay is USD $188,000 to $304,200 per year.

Some roles may be eligible for benefits and other compensation; additional benefit and pay information is available on the posting page.

The position stays open for at least 5 days, and applications are accepted on an ongoing basis until it is filled.

Benefits

• Some roles may be eligible for benefits and other compensation.