CS @ Columbia · classical pianist · I build at the intersection of music, applied cryptography, and audio ML.
- 🔒 Research (first author): Balancing Privacy and Efficiency: Music IR via Additive Homomorphic Encryption — arXiv:2508.07044. Searching and matching music embeddings without exposing them.
- 🎧 Building: forensic audio provenance, generative audio, and privacy-preserving retrieval systems.
- 🎹 Also: MTNA National Winner + Ebony Prize; NPR From the Top finalist.
| Project | What it is | Stack |
|---|---|---|
| provenance.fm | Forensic watchdog that detects when artists' music trained AI models (Suno/Udio) — ensemble fingerprinting + signed audit receipts | FastAPI · React · MERT · FAISS/pgvector |
| homomorphic | Encrypted similarity search over music embeddings (the paper's system) | Python · TenSEAL/CKKS · YAMNet · Milvus |
| aurora-sonify | Real-time pipeline turning NOAA aurora data into 30s soundscapes | Python · NOAA API · synthesis |
Python · Java · React · FastAPI · Postgres / pgvector · FAISS · Redis · TenSEAL · PyTorch · Web Audio API
Interests: music information retrieval, applied cryptography, audio ML, generative audio.