Hi, I'm Michael Xu

Senior Research Engineer at Microsoft Research - NLP Group

Pioneering AI innovations in gaming with expertise in LLMs, RLHF, and multimodal systems. M.S. in Computer Science from UIUC and experience delivering impactful software solutions at scale.

Experience

Senior Research Engineer

Microsoft Research, San FranciscoMay 2022 - Present

  • Led LLM integration across Xbox game studios, enhancing narrative complexity and player interactions
  • Delivered executive-level demonstrations of gaming AI innovations, effectively driving strategic decision-making at Microsoft Research and Xbox.
  • Worked on the Gaming Copilot Companion with on-device multimodal vision models for real-time gameplay assistance
  • Built scalable data collection platform processing 50,000+ daily human preference data points
  • Improved quest completion rates by 25% through goal-oriented conversational prompting framework

Software Engineer

Microsoft, RedmondFeb 2021 - May 2022

  • Migrated mission-critical CMS handling 27,000+ requests/second with 99.99% uptime
  • Implemented multi-region deployment architecture reducing deployment times by 60%
  • Automated log aggregation and SSL certificate management, reducing manual ops by 80%

Featured Projects

Advanced conversational AI agent for Minecraft combining LLMs and contextual RAG prompting. Most recently featured by Microsoft CEO at Build 2024, originally featured by Microsoft CTO at Build 2022.

MinecraftLLMsRAG

Blockchain-based gaming protocol using Ethereum and State Channels, with 90% reduced transaction costs

TypeScriptVueSolidity

Publications

Collaborative Quest Completion with LLM-driven Non-Player Characters in Minecraft

A novel framework for improving quest scenario completion rates through goal-oriented dialogue tracks and contextual prompting.

2024Paper

Player-Driven Emergence in LLM-Driven Game Narrative

A text-adventure game showing that players, through their interactions with the game and multiple LLM powered NPCs, can drive emergent narrative outcomes that are not pre-determined by the game designer.

2024Paper

GENEVA: GENErating and Visualizing branching narratives using LLMs

A prototype tool for generating and visualizing branching narratives. Leveraging LLMs to generate detailed narrative graphs with branching and reconverging storylines based on high-level descriptions.

2023Paper

Craft an Iron Sword: Dynamically Generating Interactive Game Characters by Prompting LLMs Tuned on Code

A novel interactive agent in Minecraft capable of rich dialogue, high-level reasoning, and complex code-generation, powered by a large language model tuned on code.

2022Paper

Education

University of Illinois at Urbana-Champaign

M.S. in Computer Science, Data Science Concentration

GPA: 3.91 • Jan 2022 - Dec 2023

Arizona State University

B.S. in Computer Science, Minor in Statistics

GPA: 4.0 • Aug 2017 - Dec 2020

Technical Expertise

Research Areas

Conversational AILarge Language ModelsRLHFMultimodal AI SystemsModel Optimization & Deployment

Technologies

PythonTypeScriptPyTorchvLLMTransformersTensorRTDeepSpeedKubernetesDockerTriton

Get in Touch

Interested in AI research collaboration or gaming innovation? Let's connect!

Contact Me