Research Engineer, Generative AI - AI Frontiers: Microsoft (Microsoft Research)

Description
About the Role
You will join the AI Frontiers lab, the team responsible for famous models and projects such as Phi, Orca, AutoGen, Eureka, and Magentic-One.

This role sits at the intersection of research and engineering. You will help evolve codebases for pre-training, mid-training, and post-training, while tackling challenges in reasoning and agentic AI. The lab maintains an open publication policy, allowing you to contribute to the academic community while shipping technology to billions of users.

Research Areas
Pre-training: Language, action, and multimodal models.

Mid-training: Long context extension.

Post-training: Instruction tuning and Reinforcement Learning from Feedback (RLHF).

Reasoning: Scaling inference time compute via Reinforcement Learning.

Action Models: Web agents and computer use models.

Orchestration: Multi-agent systems with human feedback (e.g., AutoGen).

Responsibilities
Design & Execute: Design, develop, and implement technology research projects in collaboration with researchers and product groups.

Innovation: Play a crucial role in developing and improving Large Language Models (LLMs), reasoning capabilities, and agentic AI.

Engineering: Debug and profile distributed code and optimize model architecture.

Qualifications
Required
Education/Experience:

Bachelor's + 2 years experience.

OR Master's + 1 year experience.

OR Doctorate.

Technical: 1+ years of experience developing with Python and PyTorch or JAX.

Preferred
Optimization: Familiarity with architecture and optimizations for LLMs and distributed PyTorch.

Low-level: Basic understanding of CUDA kernels.

Pipelines: Familiarity with pre-training, mid-training, or post-training pipelines.

RL: Foundational understanding of Reinforcement Learning.

Tooling: Experience with verl, Ray, Megatron, and/or vLLM is a significant plus.