Reinforcement Learning for Vision-Language Models (AI)

Location: Mountain View, California, United States

Base pay: USD $119,800–$234,700 per year across the U.S. for Software Engineering IC4; USD $160,200–$261,000 in the San Francisco Bay Area and New York City metropolitan area. Software Engineering IC5 ranges are USD $142,800–$274,800 across the U.S. and USD $188,000–$304,200 in those specific locations.

Company

Microsoft AI develops frontier artificial intelligence and multimodal model capabilities.

What you will do

• Advance visual understanding and multimodal reasoning through reinforcement learning.
• Develop targeted training tasks and synthetic data to address model capability gaps.
• Design and validate rewards, verifiers, rubrics, and model-based judges for visual tasks while detecting and mitigating reward hacking.
• Build evaluations, run controlled experiments and ablations, and perform detailed failure analysis.
• Collaborate with reinforcement learning, multimodal modeling, data, infrastructure, evaluation, scaling, and safety teams.

Requirements

• Bachelor’s degree in Computer Science, Machine Learning, Mathematics, or a related technical field with 4+ years of relevant research or engineering experience; or a related master’s degree with 2+ years; or a PhD in a related field.
• Experience developing, training, or evaluating deep learning models, with depth in at least one of model post-training, reinforcement learning, or vision-language modeling.
• Ability to formulate hypotheses, design informative experiments, analyze results, and debug model or training behavior.
• Ability to own technical projects end to end and collaborate across research and engineering disciplines.
• Proficient programming skills in Python and PyTorch or a similar framework.

Nice to have

• Bachelor’s degree with 8+ years of relevant experience; or a master’s degree with 6+ years; or a PhD with 4+ years of relevant research or engineering experience.
• Experience training models with reinforcement learning across distributed environments.
• Experience building models for visual grounding, document and chart understanding, image-to-code generation, and spatial or physical reasoning.
• Experience developing synthetic data, reward models, automated verifiers, or model-based evaluation systems and assessing their reliability and failure modes.
• Meaningful research or engineering contributions through publications, open-source work, deployed systems, or substantial technical projects.

Culture & Benefits

• Hands-on research and engineering work across reinforcement learning, multimodal modeling, data, infrastructure, evaluation, and safety.
• Benefits and other compensation may be available.
• Equal opportunity employment and reasonable accommodations are available during the application process.

Hiring process

• The position will remain open for a minimum of 5 days and applications are accepted on an ongoing basis until the position is filled.