Microsoft Research - AI Frontiers
Research Intern | May 2025 - August 2025 | Redmond, WA
Overview
During my internship at Microsoft Research, I worked with Hussein Mozannar, Adam Fourney, and Gagan Bansal at the AI Frontiers group, focusing on advancing human-centered AI agent systems.
Key Contributions
- Magentic-UI Development: Contributed to Magentic-UI, an open-source human-centered interface powered by a multi-agent system with over 7,000 GitHub stars. This platform enables human-in-the-loop agentic systems that combine human oversight with AI efficiency.
- SentinelSteps Innovation: Introduced and implemented the concept of SentinelSteps, a novel plan orchestration approach that allows agents to run longer by enabling continuous monitoring and dynamic task execution.
- SentinelBench Creation: Developed SentinelBench, a 24-task benchmark specifically designed to test web agent capabilities on long-running monitoring tasks, addressing a critical gap in evaluating dynamic, persistent agent behaviors.
Research Impact
This work advances the field of AI agents by addressing key challenges in long-running task execution and human-agent collaboration. SentinelSteps and SentinelBench provide new tools and methodologies for developing and evaluating more capable, reliable, and human-centered agent systems.