Software Engineering undergraduate · AI4SE · Building tooling for AI coding agents
I build the parts of AI coding agents that are easy to skip and hard to get right — process isolation, permission flows, failure recovery, and tests that run on the machines people actually use.
Run the official opencode on Android, inside a bundled proot + Ubuntu container. No root, no Termux. Native chat UI plus a device shell bridge.
Kotlin Jetpack Compose proot Shizuku
Drive a local opencode session from WeChat — session control, approvals, human-in-the-loop questions, file attachments and proactive pushes. One self-healing poller; CI green on Linux and Windows.
TypeScript Bun WeChat iLink
Personal blog, forked from du2333/flare-stack-blog with 89 commits on top — security response headers, first-load performance, hydration fixes, a monthly Cloudflare usage report in CI, and Android TWA support.
React 19 TanStack Start Cloudflare Workers D1
- AI4SE — how coding agents behave on real code, not in demos
- Testing discipline — 360+ tests across two projects, CI on every push
- Reliability — the boring failure modes are the interesting ones
TypeScript Bun Kotlin Compose Python React 19 Workers D1 GitHub Actions Git
