RAG-based chatbot that answers questions about UC Berkeley courses, enrollment deadlines, and campus resources. Built with LlamaIndex for retrieval, GPT-4o for generation, and Netlify for serverless deployment. Used by 400+ students during pilot.
Interactive data visualization tool that aggregates 8 years of UC Berkeley grade distributions (2M+ records). Built with D3.js and a FastAPI backend, it lets students compare grade curves across departments, instructors, and semesters before enrollment.
End-to-end NLP pipeline that fine-tunes a DistilBERT model on Amazon product reviews and serves real-time sentiment analysis via a streaming API. Processes 2,000 reviews/second with 91.4% accuracy on held-out test data. Includes a live demo UI.
Most ML papers throw math at you and call it an explanation. Here's how I actually developed intuition for why constraining weight updates to a low-rank subspace makes fine-tuning so efficient.
An OS course sounds irrelevant to application-level engineering. It isn't. Here's the mental model I built that changed how I think about every concurrent system I touch.
Open to internships, research collaborations, and conversations about machine learning, distributed systems, or anything data.
Say Hello