Senior Software Engineer, Microsoft
Sep, 2020 - Present
Designed and launched MARS (Multi-Agent Runtime Service), a cloud-native orchestration platform enabling autonomous agent coordination via group-chat architecture — defining the agentic cloud strategy for Microsoft's AI product suite.
Built a production Copilot agent for Microsoft Planner using Copilot Studio, integrating LLM-driven query, planning, and auto-reporting — the team's flagship agentic cloud use case.
Architected and stood up an AI document-generation platform for Microsoft Meetings on Azure AI Foundry — signal extraction, LLM synthesis, and structured output at scale — with an evaluation and quality-improvement pipeline (prompt tuning, grounding, and output-quality metrics) to systematically raise generated-document quality.
Contributed to Microsoft's Sydney / Declarative Agent ecosystem — plugin discovery, dynamic plugin selection (SKDS), tool invocation, and agent orchestration — applying plugin, function-calling, and orchestration patterns analogous to Semantic Kernel / Microsoft Agent Framework.
Defined the interoperability model for third-party AI agents across Microsoft 365, enabling external agents to act within Planner via the Teams Store ecosystem; designed agent permission scoping and access boundaries aligned with enterprise security and governance standards.
Defined architecture standards and reusable frameworks for enterprise agent solutions, ensuring performance, reliability, and governance across AI initiatives.
Led and coached a team of 8 engineers across distributed, onshore–offshore locations; owned technical roadmap, hiring, delivery, and engineering quality.
Managed Whiteboard–Teams integration serving 54% MAU; improved service reliability from 96.4% to 99.1% through architectural hardening and observability improvements.
Led UWP-to-web platform migration, eliminating platform fragmentation and reducing maintenance overhead across two codebases.