At Dezerv, I co-built an in-house scheduling platform that replaced Calendly for 360+ RMs, with 187ms P95, zero booking errors, and $52K/year saved.
I designed its slot assignment engine with four strategy-pattern algorithms and progressive candidate expansion. I also built a leave-computation system and an org-wide Go retry/timeout library using generics.
I shipped end to end with Flutter and React alongside Go services. My backend work focuses on microservices, event-driven systems, and concurrency.
I'm exploring how to build useful AI systems end to end, including agents with tool use and RAG, MCP, and the evaluation and guardrails needed to trust them in production. As a Research Intern at The University of Tokyo, I applied Graph Neural Networks to ASIC placement and routing in the OpenLANE chip-design flow.

