Open to Frontend / Full-Stack roles · new grad through mid-level
Software Engineer · Frontend & Full-Stack
I build production web apps end to end — React/TypeScript front ends, Node/Firebase/serverless back ends, and AI/LLM integration in between.
See my work// selected work
Please pick one and click around.
Pet-Centric Social Platform
A full social app for pet people: posts and feeds, multi-pet profiles, check-ins at pet-friendly places, in-person meetups, notifications, and moderation. It runs on React 19 with a callable-first Firebase backend — Cloud Functions and Firestore security rules doing the heavy lifting, with Cloudinary for media and Geoapify for places.
AI Agent Session Replay
Drop a Claude Code transcript on the page and step through the run: the messages array as it grew, the step that blew up the context, and what it had read before it wrote. Compare two runs, or index every session on your machine at once — all parsed incrementally in the browser, nothing uploaded. A rule set runs the same checks from the command line and exits non-zero, so it fits in CI.
Real-Time AI Communication Coach
A real-time coach that scores how you're coming across as you speak — live transcription, LLM tone analysis, and audio signals combined into one score. It adds roleplay "sparring" personas, an explanation timeline behind every reminder, AI recaps and a zero-mic demo, all served through a Vercel serverless proxy to Groq (Whisper + Llama 3.3 70B) with a circuit-breaker fallback chain.
Brand Site & Dealer Portal
Immigration Timeline Tracker
A statically generated dashboard for immigration backlogs: US green card categories (EB-1 through EB-5 and family-based) alongside Canada's Express Entry draws. Ten years of history, per-category wait estimates, and official regulatory news — fed by three scheduled pipelines that pull from official sources only.
Visual Git History Explorer
Three views over a public repository: Replay walks the default branch commit by commit, Compare puts any two points side by side, and Insights tracks how size and structure moved. File trees are rebuilt from checkpoints and diffs, the GitHub token stays on the server, and a synthetic demo spends no API quota.
Native macOS Day Planner
Attach a reward to each task; clear them all and the app throws confetti for a
Perfect Day. It builds as a plain Swift package, SwiftUI and SwiftData only.
Two hidden flags: --snapshot renders every screen to PNG offscreen,
--selftest checks what screenshots cannot, like drag-and-drop
placement and schema backfill.
// experience
Most recently: building a shipping platform from zero.
WeShipItNow · multi-carrier shipping platform
2024 — present
E-commerce sellers compare live rates across USPS, FedEx, UPS and Amazon, buy discounted labels, and track every shipment in one place. I built the React front end from scratch against a hybrid GraphQL-and-REST API, and co-designed the backend behind it — idempotent purchase flows, a durable job queue for carrier calls, and an adapter layer that keeps four carrier APIs behind one internal contract.
Supported a graduate database course covering SQL and Microsoft SQL Server.
// teaching
Interactive courses and practice tools, from guided explanations to independent drills.
Data structures you can watch: memory diagrams, animations you can step or predict, Java/Python/JS side by side.
14 chaptersAlgorithms, frame by frame — decision trees, DP tables, shrinking intervals, plus a race that bills each sort in comparisons and moves.
13 chaptersHTTP, REST and GraphQL end to end — versioning, caching, auth — against a mock API that runs in your own browser.
12 chaptersTypeScript from first principles, with the real tsc in a Web Worker: your code, its own error codes and positions.
12 chaptersRedis through a single system: caching, concurrency guards, read models, a hands-on lab and a fault simulator.
8 stopsScaffolding taken away a step at a time: 105 recall questions, 148 exercises, 25 coding problems, 7 timed builds, six guided plans.
80 lessonsMessage history, tool calls and the agent loop — you write a working agent by stop three, then cost, context, failure and trust.
14 stopsSix real debugging cases you can break, fix, and lock down with the regression test that catches each one.
6 exhibits// lab
Browser experiments, small games and visual sketches. Click one and it runs.
// skills
The stack I reach for.
// about
I came to engineering the long way — six years as a senior data analyst before a STEM master's at Northeastern. That path stuck with me: I build with an eye on what actually ships and what the data says afterward, and I like owning a product end to end, from a responsive React front end to the serverless functions and data model behind it. Lately I'm most interested in getting the AI/LLM pieces right — and I work in both English and Chinese.