My Projects
A comprehensive collection of my work, showcasing various technologies and solutions
Trippy (LAHacks @ UCLA)
- Built a travel platform using Fetch.ai agents and OpenAI GPT APIs for semantic companion matching and smart logistics.
- Designed modular agents for travel tasks (flights, lodging, transport, etc.) using LangChain, LangGraph, and prompt engineering.
- Used RAG with Pinecone to match travelers via vectorized itineraries, enabling context-aware matches and ice-breakers.
AI Autograder
- Built a full-stack AI autograder with React (frontend), Django + PostgreSQL (backend), and FastAPI for modular LLM services.
- Integrated GPT-4 via LangChain and RAG pipelines for rubric-based grading, semantic answer-key matching, and personalized feedback.
- Applied spaCy, scikit-learn, and sentence-transformers for text parsing, vectorization, and semantic similarity scoring.
- Used FAISS and Redis for fast vector search and caching, enabling real-time grading and dynamic document retrieval.
Spartan MyCompanion
- Developed a full-stack student event networking portal hosted on AWS Amplify, supporting 500+ concurrent users, with automated testing and deployment via GitHub Actions.
- Integrated AWS API Gateway, Lambda and DynamoDB sink for functionalities like event posting, replies, and voting, with real-time chat rooms using WebSocket API and event notifications using fanout with AWS SNS.
- Introduced infrastructure as code with AWS CloudFormation and JSON templates for automated deployment.
Rideshare Cost Optimization Tool
- Developed a web application that cut time-based pricing for Bay Wheels by 30% by utilizing Java, Spring, and PostgreSQL, along with advanced data structures and refactoring techniques for improved code quality.
- Employed Dynamic Programming, Dijkstra's, and Yen's algorithms to analyze and compute the most cost-effective paths for long journeys.
- Developed an intuitive dashboard using React for route visualizations.
5G Handover Prediction System
- Engineered a machine learning pipeline using TensorFlow, Scikit-learn, Pandas, and NumPy to process 103K+ 5G network samples.
- Deployed an LSTM signal predictor and a CNN-BiLSTM hybrid classifier to predict cellular handover trigger events.
- Resolved a 47:1 class imbalance using Temporal SMOTE with 50-timestep sliding windows, achieving 98.37% prediction accuracy.