I'm pursuing a BEng in Computer Science at HKUST, with a minor in Actuarial Mathematics and expected graduation in 2028. I build AI-powered tools and full-stack products that make everyday tasks easier.
Most recently, I worked as a Software Engineer at Sun Life Financial, automating financial workflows and building AI-assisted document review tools. Before that, I worked on automated market making at BNP Paribas, developing Python and C++ trading automation and an internal AI chatbot for document retrieval.
I'm also the founder of Lucida, an assistive technology venture building affordable AI-powered navigation for people with visual impairments. Backed by a $100K HKSTP Ideation Grant and seven NGO partnerships, Lucida has enabled beta testing with 25+ users.
A 2-vs-2 Rajasthani Canasta game with online rooms, solo play against bots, and a dedicated rules engine for melds, wild cards, team reserves, and round scoring.
Provider-neutral runtime foundations: typed contracts, validated run-state transitions, scoped default-deny policies, and a local policy-validation API, backed by property and adversarial tests.
An evidence-first dependency upgrade workflow: inspect a Python repo, approve a migration plan, verify repairs in Docker, and review the patch with check logs and independent evaluation.
A daily dashboard combining live markets, curated news, weather, job listings, and personal planning. Optional Gemini briefings synthesize the day's signals, with a rule-based fallback.
A shared coding-agent engine across a VS Code extension, browser dashboard, and CLI. Route tasks across model providers, review proposed edits, approve tests, and inspect local execution traces.
A Python research pipeline combining XGBoost, LightGBM, and LSTM signals with walk-forward validation, out-of-sample backtests, Monte Carlo simulations, and layered risk controls.
A 3-layer real-time system: KNN/SVM classifier over 89-dim MediaPipe hand landmarks, disambiguated by a trigram language model, confirmed via TTS + gesture recognition.
Full-stack home inventory app with recursive location trees, QR/barcode batch scanning, real-time Supabase sync, force-directed knowledge graph, and Stripe subscription tiers — published on iOS & Android.
Pick a location on a map and Atlas ingests live data from 6 external sources — demographics, competitors, parcels, traffic, schools — scores candidate sites, and runs an LLM bull-vs-bear debate to synthesize a recommendation.
A Next.js platform built with my team for the Morgan Stanley x Zubin Foundation hackathon — serving Hong Kong's ethnic minority community with accessible resources, translated support, and an AI chatbot layer to connect users with services.
An experimental deep-learning project using LSTM neural networks to forecast urban traffic flow from time-series sensor data. Built on Keras with Theano backend and MongoDB-stored hyperparameter sweeps for grid optimization.
A hackathon app that turns accountability into a game: your squad shares a castle whose HP degrades as members doomscroll. Set per-app time limits together and defend the castle — or watch it fall.
Next.js prototype that maps customs classification codes between China (CCC) and Indonesia (BTKI/AHTN) using GPT-4o. Search by description, HS code, or national tariff line — ranked matches with confidence, match basis, and source references.
I want to build systems that are both technically rigorous and genuinely useful to real people — whether that's quantitative models that move markets, or assistive tools that open doors for underserved communities.
I'm drawn to roles at the intersection of software engineering and intelligent systems — quant development, AI research engineering, and product roles at companies building things that last.
Download my latest documents below.