Two sets. Work files are systems I shipped inside a company, so the code is not mine to publish. Personal files are things I run in my own account — some public, most private because they hold other people's data.
6 projects. Every one is on the same stack I would pick at work: Bun, typed end to end, Postgres behind a tRPC surface.
Commissioner platform for 6 dynasty fantasy football leagues and 74+ members.
Real-time Sleeper API sync over 231K+ player-week stats from 2020 onward, a 9-phase league lifecycle state machine, magic-link auth, cross-league ROI rankings, a peer-to-peer betting and settlement workflow, and 15 transactional email templates. Daily 4am sync handles league discovery and format detection on its own.
Retail checkout automation, revived from an abandoned project and adopted by a community.
A resurrection of Bird Bot: a Python automation tool for high-demand retail drops. Picked up a dead codebase, modernized it, and maintained it in public long enough for it to become the reference fork.
Social collection platform for anime figure and manga collectors.
Collection tracking, shareable shelves, price history, and a public profile surface. Multi-source ingestion workers keep manga, figure, and pricing data current; the whole thing runs the same Bun + tRPC + Astro stack, containerized behind Caddy.
Booking platform for a restored akiya in Chiba Prefecture, Japan.
Direct-booking site for a restored Japanese home — availability, rates, and reservations without a marketplace in the middle. Same stack discipline as everything else: typed end to end, Drizzle migrations, mock-data-first UI.
Cross-game index for gacha titles: characters, banners, and rates in one place.
A monorepo experiment in modeling messy, community-sourced game data behind one typed API surface, with the ingestion workflow documented for agents to extend.
Discord bot for gacha communities: coupon redemption, reset alerts, moderation.
Automated coupon redemption, daily reset notifications across multiple games, URL embed repair, and community moderation tooling. Runs on a schedule with S3-backed state and a Vitest suite behind it.
Systems I owned inside a company. Written as case files, since the source is not public. Each one maps to a box on the schematic.
Internal platform where non-engineers describe a tool and get a deployed app.
Builder and generated apps run on ECS Fargate behind CloudFront with CI/CD, Auth0 end to end, data access brokered through MCP, and version history with rollback. Shipped the prototype and scoped the company-wide release.
Replaced ad hoc database access for AI agents with one governed, audited path.
Every agent request is authenticated, logged to an audit trail, and labeled with a data classification. Self-serve access approval takes turnaround from days to minutes.
Event-driven rearchitecture at 10x throughput with 90% faster evidence clips.
Replaced brittle legacy ingestion with an event-driven architecture, integrated AI event detection models tuned for production throughput, and rebuilt evidence clip generation for reliability downstream.
EHR interoperability for decentralized trials with Bayer, J&J, and BioMarin.
Spring Boot microservices and MuleSoft integration layers transforming partner data into FHIR, so remote and in-clinic nurses could run trials on accurate, interoperable patient records.