AI work
This section collects the projects where I have used AI tools. Each one has a write-up of what was built, how it works, and how AI was used along the way.
The thread that connects them is in AI and engineering leadership →
One engine running three evidence-first signal products — TechDrift, GamingDrift, SportsDrift — each white-labeled with its own source catalog, database, and identity. A self-hosted LLM, Laya, handles only bounded advisory work (topic clustering, keyword dedup, per-user analysis) and never gates what a user sees. A full admin surface covers sources, users, notifications, and ingestion health. Android and iOS apps are coming soon.
An engineering-manager dashboard that scores each direct report’s capacity, pace, and open pull requests as Red, Amber, or Green, and suggests a starting point for the next one on one. It runs on simulated data, and the write-up covers the scoring rules, the design decisions, and how AI was used to build it.
A print-on-demand shop run by nine AI agents, each with one job. A request moves through concept, design, copy, publishing, website sync, and promotion until the product is live for sale, with no manual hand-offs between tools. The write-up covers the agents, the workflow, and the controls that keep it dependable.
Claude agents scan a dozen-plus repositories, from backend services to front-end clients, log what they find as ranked Jira tickets, and open a pull request with a proposed fix for each, giving a team one prioritized list to review and burn down each quarter. The first pass logged more than 150 bugs. A live demo app replays the workflow on simulated findings.
A private dashboard for my lab, configured heavily with AI. Rich cards show live numbers from every service, bookmarks managed in Karakeep are synced to the bottom of the page, and a cron job keeps it current every 15 minutes. It sits behind a VPN, so the write-up uses edited screenshots.
