Lead supervised fine-tuning, preference optimization and reinforcement learning (RL). Develop reward models and task-based reward signals that improve reasoning, multi-turn task completion and tool-use correctness.
Build data and feedback pipelines with engineering teams. Curate demonstrations, preference data, synthetic examples and agent interaction histories, with clear data permissions, versioning and separation between training and evaluation data.
Establish reproducible evaluations with strong baselines and controlled comparisons. Measure generalization and safety, prevent evaluation-data leakage and detect models exploiting rewards without completing the intended task.
Lead inference experiments in quantization, distillation and speculative decoding to reduce computation and memory use. Measure time to first token, generation speed, tail latency, throughput and cost under representative load, against explicit model-quality targets.
Partner with systems engineers on distributed training, agent rollout generation, batching and caching. Reduce training GPU-hours and serving cost while preserving convergence, task quality and reliable recovery.
Take successful experiments into production through staged releases, monitoring and rollback plans. Work with engineering teams to preserve data privacy, tenant isolation, streaming behavior and token-accounting correctness.
Qualifications
Required Qualifications
Master's Degree in Computer Science, Machine Learning, Statistics, Engineering or a related technical field AND 8+ years of relevant applied science, machine learning or engineering experience; OR Bachelor's Degree in a related technical field AND 10+ years of relevant experience; OR equivalent experience.
Experience managing applied scientists or machine learning researchers, including hiring, coaching, performance management and delivery through a team.
Hands-on experience delivering both large language model post-training and inference optimizations, with measured improvements in model quality, latency or cost.
Strong Python and practical experience with PyTorch, JAX or comparable frameworks. Strong foundations in machine learning, optimization, statistics and experimental design.
Understanding of distributed training and GPU serving bottlenecks, including compute, memory and communication trade-offs.
A record of turning research into production capabilities with engineering and product partners, and explaining technical trade-offs clearly.
Preferred Qualifications
Experience with agent post-training, reward modeling, preference learning or RL systems that improve held-out task success.
Experience with training toolchains such as verl or TRL, and serving runtimes such as vLLM, SGLang or TensorRT-LLM.
Experience co-designing model improvements and serving optimizations across model families or hardware platforms.
Research, open-source or production contributions that demonstrate reproducible scientific results and practical customer impact.
This position will be open for a minimum of 5 days, with applications accepted on an ongoing basis until the position is filled.
Microsoft is an equal opportunity employer. All qualified applicants will receive consideration for employment without regard to age, ancestry, citizenship, color, family or medical care leave, gender identity or expression, genetic information, immigration status, marital status, medical condition, national origin, physical or mental disability, political affiliation, protected veteran or military status, race, ethnicity, religion, sex (including pregnancy), sexual orientation, or any other characteristic protected by applicable local laws, regulations and ordinances. If you need assistance with religious accommodations and/or a reasonable accommodation due to a disability during the application process, read more about requesting accommodations.
