Projects

Liquid Cache

Query-aware pushdown cache for DataFusion. Transcodes Parquet into a cache-only format and pushes predicates directly into storage, keeping only what each query needs in memory. A research project under the UW-Madison Systems Lab.

EngrAAV

Deep learning directed-evolution pipeline that designs recombinant AAV capsids optimized for viral yield, cutting gene therapy manufacturing costs by ~96%.

AutoFX

AI-powered Electron app that uses RAG and local LLMs to generate and run FFmpeg pipelines from natural language, automating complex media processing tasks.

Text2Synth

FPGA-based real-time audio synthesizer for zero-shot text-to-patch generation. A custom SystemVerilog accelerator optimizes FM/subtractive synthesis parameters in a joint Audio-CLIP embedding space, delivering sub-10ms inference with steady-state 48kHz audio output.