What I'm building
A small set of proof-oriented threads, product work, personal systems, and infrastructure that show how I think in public. The common thread is simple: real AI value appears when runtime, memory, and deployment hold up.
SleepHub
Ambient, bedside AI product work for autonomous sleep coaching and tracking. This is where product, device integration, sleep analysis, and real-world user loops meet. It is the clearest public signal of my applied AI work and my bias toward systems that have to earn trust over time.
Knowledge OS
A markdown-first personal operating system for memory, retrieval, and public expression. Files as source of truth, agents as readers and writers, git as history. It is both a working system and a statement about inspectability, explicit memory, and how AI should integrate with real workflows.
Agent infrastructure
Runtime correctness, memory continuity, and workflow tooling for human-AI collaboration. Multi-agent orchestration, session handling, tool contracts, and guardrails encoded into the environment itself. The interesting work is what the system does when context is partial, state drifts, or humans interrupt.
Patient–family record sharing
A lightweight digital health layer for the period after a hospital visit. The patient records what they were told — appointments, instructions, preparation items — and chooses to share those notes with a family member or care companion they designate, so both sides can stay on the same page about what comes next. Scoped deliberately to patient-led record sharing, preparation, and communication support outside the hospital — not medical judgment, not triage. Early stage; being shaped with real patient and family input before any code commits to a direction.
Open Wall
Multi-angle indoor room reconstruction powered by OpenAI
gpt-image-2. An early experiment in turning image models
into a spatial reconstruction workflow rather than a one-off image generator.
Workflow tools
Small public tools around real workflows: a Toss Securities CLI, a Google Timeline visualizer, and dotfiles for environment management. These are deliberately practical systems: narrow enough to inspect, useful enough to keep.