From product ideas
to an AI platform

Micro brings email, CRM, meetings, tasks, and documents into one workspace, giving people and AI agents the context to act across them.

Most features began as a conversation or rough idea, without a written spec or design. I turned those ideas into a defined problem, a technical design, and a delivery plan. I chose the stack and system boundaries, worked out what was feasible, and split delivery into phases that matched the team’s bandwidth.

As Head of Engineering, I led the other engineers while staying hands-on across the interface, APIs, and background infrastructure. I stayed responsible for how the pieces worked together: permissions, retries, persisted state, and the experience of finding the result.

An assistant needed to distinguish what matters right now, what happened before, and what should still matter next week.

I conceived, designed, and built a layered memory system inspired by how people remember: working memory for current priorities and open loops, episodic recall of past conversations, and long-term memory for durable facts and preferences.

Result The agent could carry context across chats and automations, retrieve earlier details, and update durable memories when preferences changed.

Important details lived inside conversations; keeping structured records current required manual work.

I built the shared record model and autofill pipeline connecting email and meeting context to typed properties, relationships, and workspace permissions.

Result 1,062,093 distinct production autofill runs, including 266,899 during September 20–October 19, 2026.

A scheduled agent needed to gather context, call tools, save its work, and make the result available later. Failures could happen between any of those steps.

I built scheduling, background authentication, execution, and run-history controls, then introduced durable stages for preflight, agent work, persistence, and finalization.

Result 21,161 distinct successful automation runs during August 10–October 19, 2026; 12,923 during September 20–October 19.

A recording only became useful when its transcript and follow-up work reached the right place.

I built meeting recording across calendar-driven bot creation, API controls, provider webhooks, and the meeting interface. Transcripts fed summaries, follow-up tasks, and CRM autofill.

Result 3,346 retained completed meeting recordings with recording links, rather than scheduled joins or bot attempts.

Agent actions and generated apps needed access to the same context, permissions, and persisted state as the product.

I built and extended agent tools, reusable skills, and MCP access, and led prompt-to-app development across the agent, APIs, and web product.

Result 1,105,198 recorded tool calls and 161,122 retained AI conversation messages across the agent platform.

TypeScript/JavaScript, React/Next.js, Node.js/Express, PostgreSQL, AWS workers and queues, Inngest, Mastra, Vercel AI SDK, Anthropic/OpenRouter, Recall, Daytona, Yjs/Hocuspocus, SwiftUI, and Tauri.

I learned to turn open-ended product ideas into systems people could use, then keep owning the details as the team and product grew.