Mike Ichikawa

I shipped Microclaw solo from an empty repository to a live, transactable Microsoft Commercial Marketplace AI agent for Microsoft 365. It connects 85 function-calling tools across 12 Microsoft 365 services through a plan/act/observe loop.

I built the retrieval layer that determines which tools the agent can call, improving retrieval recall from 92.7% to 95.0% against an embedding-only baseline. I also delivered RAG semantic memory, model routing with mid-task escalation, failure recovery, release-gating evaluations, Microsoft AppSource certification, and billable Fulfillment and Metering API integration.

I built a Python multi-agent workflow engine from scratch that converts plain-English requests into validated control-flow graphs of single-purpose agents. Workflows require human approval before acting, reuse a shared agent library, and improve through corrected examples rather than fine-tuning.

Before moving into agentic systems, I spent eight years teaching undergraduate mathematics and three years in Intel semiconductor R&D, where I worked on Ivy Bridge physical design and built automation for physical-verification workflows. I bring that foundation in mathematics, evaluation, and systems reliability to AI-native product development.