Computer Science @ Langara College | Full-stack Software Engineer

Open to: Software engineering internships & new-grad roles (full-stack, backend, AI-adjacent product engineering)

I own products end-to-end — UI, APIs, data models, auth, automated tests, CI, and production deploys.

3 live applications on Vercel · 113+ automated tests · CI on every change.

Open source: typescript-eslint — 2 merged PRs on ESLint's TypeScript plugin (unified-signatures, no-meaningless-void-operator).

Turned unstructured customer feedback into actionable, shareable reports, measured by a full production upload → analyze → export workflow with 51 passing unit tests and GitHub Actions CI, by building a Next.js/TypeScript application with OpenAI-driven theme and sentiment analysis, Prisma/PostgreSQL persistence, PDF/CSV export, and password-protected share links.

Automated the improve-and-ship loop for Python repositories, measured by a 5-phase async pipeline (clone → analyze → LLM refactor → validate → GitHub PR) with 5 safety gates, 4 static analysis tools, and 35 pytest tests, by engineering FastAPI + Celery workers on PostgreSQL/Redis, Claude-powered refactors with 3 retry attempts per issue, and a Next.js operations dashboard deployed across 6 services (Railway + Vercel).

Improved search quality when exact keywords miss the intent, measured by a 3-stage retrieval pipeline (vector → semantic → keyword fallback) with 23 unit + 4 E2E tests and streaming LLM responses in production, by implementing Pinecone embeddings, Next.js APIs, authentication, and usage analytics.

• Raised release confidence across 3 production apps, measured by 113+ automated tests and CI on every push, by writing unit, API, and E2E coverage around core user and pipeline flows.
• Shortened time-to-insight for review data, measured by one-click CSV upload → themed summary in a single session, by designing async analysis pipelines and persistent report storage.
• Prevented unsafe LLM output from shipping, measured by 5 validation gates blocking PRs unless syntax, lint, security, diff, and pytest checks pass, by treating model output as untrusted until structurally verified.
• Reduced irrelevant RAG answers, measured by layered fallback retrieval when vector search underperforms, by combining Pinecone similarity, semantic re-ranking, and keyword search.
• Kept long-running work off the request thread, measured by sub-second API responses while clone/analyze/refactor runs for minutes, by offloading pipelines to Celery workers with live task polling.

TypeScript Typed web code

JavaScript Web interactivity

Python Scripts & data

Java Backend apps

C++ Performance

Next.js Full-stack web

React User interfaces

Node.js Server & APIs

PostgreSQL App database

AWS Cloud hosting

Vercel Live deploys

Git Code history

Terraform Cloud setup

Also in my projects: Prisma · OpenAI API · SQLite · MUI · Tailwind CSS · Playwright · Vitest · GitHub Actions