Projects
KONSEI Skin Lab
↓Deep learning model trained to identify and classify skin conditions from images. A convolutional neural network analyzes user-submitted photos and returns a diagnosis with confidence scores. Paired with a mobile app for on-the-go skin health tracking and condition history logging.
Dormly
↓Full-stack campus marketplace with WebSocket-powered live listings and 1-on-1 chat via Socket.io. Features JWT authentication with .edu email verification to restrict access to verified students, plus a reputation and ratings system for buyers and sellers.
Drifty Studio
↓Desktop songwriting tool powered by an LLM that acts as a ghost collaborator. Feed it a mood, theme, or rough lyric and it generates verses, hooks, and bridges that match your voice. Designed to remove creative blocks while keeping the finished song authentically yours.
(PinDex)
Cross-platform mobile app built with Expo/React Native for real-time bowling shot tracking and session analytics. Integrates a Groq LLM (Llama 3.3-70b) to generate personalized AI coaching feedback, and a curated ball database with AI-assisted lookup for equipment recommendations.
NexTrip
↓Enter a destination, budget, and interests and NexTrip generates a complete day-by-day itinerary with curated activities, timing, and smart fallback suggestions for any gaps. Built with Python and Flask, the backend dynamically assembles trip plans and handles edge cases to ensure every schedule is coherent and ready to follow.
(BoomBoom!)
Real-time multiplayer Jeopardy-style game built for classroom engagement. Students join a live session from their phones, which act as buzzers. First to tap wins the point. A host screen displays questions and a live leaderboard while Flask-SocketIO syncs buzz-ins across all devices in milliseconds, handling race conditions to crown a single winner per round. Players earn and manage power-ups through a custom inventory system, adding a strategic layer to every round.
Digit Recognizer
↓Convolutional neural network trained on the MNIST dataset (60,000 images) achieving 99.1% test accuracy. A Flask API serves the model, and a browser canvas lets you draw any digit and get an instant prediction with confidence scores. Preprocesses drawn input to match the 28×28 grayscale format the model expects.