CS @ Duke (minor in Machine Learning & AI), building and evaluating LLM systems. I like taking an AI feature end-to-end, building it, proving it works, and improving the model behind it. Targeting Summer 2027 AI/ML internships.
Currently: model-evaluation & alignment work at Duke Code+ Program (LLM-as-judge + execution-based scoring, QLoRA-DPO) and sickle-cell-disease data research at the Duke Global Health Institute.
Tools I reach for: Python · PyTorch · Hugging Face (TRL/PEFT) · FastAPI · LangGraph · vLLM/Slurm · Docker · TypeScript/Next.js
Featured projects
- lyrics_mood_predictor (live demo): fine-tuned DistilBERT for song-lyric mood prediction, served with FastAPI + ONNX, with SHAP explanations and Qdrant vector search over 76k songs.
- trading-agent-analyzer (live demo): a multi-agent LLM stock analyzer (Streamlit UI built on the open-source TradingAgents / LangGraph engine) that produces comparative buy/sell/hold rationales.
- ddl_reminder (live app): login-gated; demo GIF in the repo) — Next.js + Supabase deadline tracker I use daily: natural-language entry, email reminders, Apple Calendar feed, Canvas/Gradescope sync, installable PWA.

