Saravan Rajmohan
Saravan Rajmohan is a Microsoft AI research executive who leads applied research on AI systems and agents. He coauthored TaskWeaver in 2023, a code-first framework for large language model agents, and UFO in 2024, an agent for operating Windows applications. His research later covered how language models use memory through Minerva in 2025 and how agents communicate through learned message formats in 2026. As of October 2026, he is a Vice President and Distinguished Engineer for AI and Applied Research at Microsoft and leads M365 Research.
Timeline
-
2023
TaskWeaver's code-first framework
In 2023, he coauthors TaskWeaver, a code-first framework for large language model agents that turns user requests into executable code and treats user-defined plugins as callable functions.
-
February 2024
Leading M365 Research
In February 2024, he leads Microsoft's M365 Research group as a Partner Director, focusing its work on AI and applied research.
-
2024
UFO for Windows
In 2024, he coauthors UFO, a GPT-Vision agent that uses a dual-agent framework to operate Windows applications and handle tasks across multiple apps.
-
2025
Minerva memory benchmark
In 2025, he coauthors Minerva, a framework that generates tests to measure how language models use memory, including tasks such as searching, recalling, editing, and maintaining state.
-
2026
Learning agent communication
In 2026, he coauthors OPTiMACS, which learns message representations for agent communication and evaluates them on collaborative problem-solving tasks.
-
As of October 2026
Vice President and Distinguished Engineer
As of October 2026, he is Microsoft's Vice President and Distinguished Engineer for AI and Applied Research, leading M365 Research under M365 Core with a team of more than 40 researchers.
Sources and credits
Photo: We have no published photo of Saravan Rajmohan yet. Send us a free-licensed photo.
- TaskWeaver: A Code-First Agent Framework
- [CDS Seminar] Systems+AI Research @ Microsoft: Department of Computational and Data Sciences- Indian Institute of Science, Bangalore
- UFO: A UI-Focused Agent for Windows OS Interaction
- Minerva: A Programmable Memory Test Benchmark for Language Models
- Learning Optimal Message Representations for Agentic Communication
- Saravan Rajmohan at Microsoft Research