Syllabyss converts uploaded course documents (PDFs, PowerPoints, Word files) into interactive study modules. Six game modes — Dive, Apogee, Leap, Pairs, Blitz, and Arena — are generated from the material using Gemini, with every answer traced back to the source page. Sonar, an AI coach powered by Claude Sonnet and LangGraph, applies Bayesian Knowledge Tracing to identify gaps and recommend targeted resources.

• Delivered 40+ features in one weekend across six polished game modes.
• Every generated answer references its exact source page in the uploaded material.
• Sonar uses Bayesian Knowledge Tracing — mastery is evidence-based, not LLM speculation.

Developed input parsing logic and processed structured JSON data for backend integration into Java classes. Leveraged Spring Boot, Python, and FastAPI to build and connect backend services across a dockerized full-stack system.

• Converted structured TAC inputs into backend-ready Java models.
• Integrated multiple backend services across a containerized system.
• Improved reliability of data handoff between parsing and fitting workflows.

Led the frontend for a responsive activity planner that combines OpenWeather data, Google Places recommendations, and a focused Tailwind CSS interface to suggest rainy day plans.

• Owned the frontend experience from layout to interaction polish.
• Connected weather context to location-aware activity planning.
• Built a clear mobile-first flow for fast decision making.

Built LuxBot by integrating computer-vision emotion detection with multimodal AI to enable real-time therapeutic conversations. Added dynamic front-end components, including an animated motivational banner, to improve engagement and responsiveness.

• Integrated live facial-expression recognition with natural-language responses.
• Reduced interaction friction through adaptive emotional context.
• Enhanced engagement with animated interface feedback.

Created a Python-based genre classifier using data analysis workflows and KNN classification to study audio feature patterns and evaluate prediction quality.

• Built and evaluated a KNN classification pipeline.
• Explored feature relationships across genre-labeled audio data.
• Documented model behavior with analysis-driven findings.