Self-directed builds, prototypes, and technical case studies.
Personal Projects
NASA Space Apps 2025 prototype for exploring 607 space-biology publications. Built by a six-person team, it combines PMC metadata, tag filtering, timeline and subject-tree views, and a Flask/OpenAI summarizer.
Next.js / React / Material UI / Flask / OpenAI / BeautifulSoup
read the case study ->Serverless Discord bot for market and macro data. It uses Discord slash commands, AWS Lambda, DynamoDB caching, Massive, Binance, Marketaux, SEC RSS, and Federal Reserve feeds without running an always-on gateway process.
Python / AWS Lambda / DynamoDB / Discord API / AWS SAM / Matplotlib
read the case study ->Expo/React Native finance app that keeps core budgeting data on-device. It combines SQLite/Drizzle storage, receipt-backed expenses, subscription tracking, loan scenario modeling, backup/restore, and optional AI insights through a backend proxy.
Expo / React Native / SQLite / Drizzle / Expo Router / Jest
read the case study ->Two-player, terminal-based card game written in C++20 for a CS246 final project. The engine models minions, enchantments, rituals, and spells over a shared card/target hierarchy, using the Decorator, Observer, Strategy, and Factory patterns so new cards are data plus small classes instead of edits to the game loop.
C++20 / OOP design patterns / Make / g++
read the case study ->Always-on Python service that watches Reddit chatter and Hyperliquid perp flow for unusual activity. It normalizes two very different sources into one event schema, scores them with engagement and z-score/whale heuristics, dedupes, and pushes only threshold-clearing alerts to a Discord channel, packaged in Docker.
Python / asyncio / WebSockets / PRAW / Docker / Discord API
read the case study ->Python desktop utility for Honkai: Star Rail account chores. It wraps HoYoLab endpoints in a CustomTkinter app for stamina tracking, daily check-in, character browsing, code redemption, scheduled refresh, local caching, and a one-file Windows build.
Python / CustomTkinter / Requests / Schedule / PyInstaller
read the case study ->University of Waterloo WiM Directed Reading Program. A three-person mentored project that builds a shared-bottom multi-task neural network to forecast several assets jointly, benchmarks it against linear regression and single-task neural nets, and finds that multitask learning bought efficiency rather than accuracy on noisy next-day prices.
Python / TensorFlow / Keras / scikit-learn / pandas / yfinance / Matplotlib
read the case study ->