Building things that matter.

A Python productivity tool that uses real-time webcam gaze tracking to detect when you look away from your screen — then punishes you by blasting a full-screen meme or brainrot clip. Built with MediaPipe, OpenCV, PyQt6, and python-vlc. Features Chaos Mode, a grace-period warning system, and a configurable media library.

• Built the full gaze detection & CV pipeline using MediaPipe Tasks API with adaptive smoothing and per-session calibration
• Resolved a camera fake-focus race condition and media looping bug affecting overlay timing
• Integrated a full CI/CD pipeline with pytest and GitHub Actions — headless-safe test suite across 6 modules
• Collaborated with a 4-person team; led CV integration and coordinated with PyQt6 frontend

A Streamlit dashboard for Super Smash Bros. Ultimate players. Analyze character matchups, find counterpicks with stat radar charts, track learning progress from beginner to pro, and browse the full tier list. Deployed via Streamlit Community Cloud and optimized for mobile.

• Developed a responsive web application for Super Smash Bros. Ultimate players to analyze matchup spreads and track progression across 80+ playable characters
• Built an interactive counterpick engine with per-matchup stat radar charts to help players identify optimal character choices
• Designed a structured learning tracker mapping skill milestones from beginner to pro across all characters
• Integrated a full tier list browser with filtering and sorting by stats, tier, and matchup data

A Streamlit-based dashboard for analyzing and backtesting algorithmic trading strategies. Features interactive equity curve visualizations, a configurable strategy builder, and SQL-powered performance reporting via Flask and SQLAlchemy. Supports drawdown analysis, trade history, and complex SQL queries across stored backtest results.

• Engineered a financial metrics calculator supporting 17+ distinct performance indicators including Sharpe ratios and max drawdowns in an open-source market analyzer by processing backtest equity curves and trade logs.
• Validated end-to-end backtesting pipeline reliability by achieving full metric accuracy across 5 distinct trading strategies
• Deployed as a full-stack Streamlit + Flask app with Plotly visualizations

A multimodal fake news detection web app built with Streamlit. Classifies misinformation in text articles and videos using DistilBERT, CLIP, and Whisper. Includes article scraping and video transcription.

• Trained a multimodal classification pipeline combining DistilBERT for text, CLIP for images, and Whisper for video transcription
• Built an article scraper and video transcription module to support real-world input from URLs
• Integrated all three modalities into a unified Streamlit interface with confidence scores per modality
• Handled edge cases for text-only, image-only, and video inputs with graceful fallback logic