Thomas Cheng Li

Developer debugging tool that records screen activity via a single hotkey and uses AI-powered temporal video analysis to identify visual bugs frame by frame. Outputs structured diagnostics directly to AI coding assistants like Claude Code and Cursor. Available on macOS, Windows, and Linux with MCP server integration.

A genomic database aggregating 9B SNVs, 80M indels, and 160+ annotations to link disease traits to genetic variation, adopted by 20+ labs. Built FAVOR-GPT, a RAG-powered assistant for natural-language database querying. Applied PCA and embedding models for large-scale similarity search; improved runtime by 80% via GiST-based indexing. Full-stack app with PostgreSQL, FastAPI, and React.

Designed a compact 2-layer infotainment PCB (60% smaller) with full I/O and direct Raspberry Pi 5 interface. Performed SMT reflow, debugging, and continuity/ESD testing. Developed STM32 firmware in C/Python for peripheral communication, logging, and UI control over serial and I²C. Integrated Pi 5 touchscreen with wireless connectivity and designed CAD dashboard mounts.

Automated waste sorting system using Vision Transformers (ViTs) with self-supervised DINO pretraining on the RealWaste dataset to classify recyclable materials. Integrated a Raspberry Pi for real-time image processing and Arduino-controlled servo motors for mechanical sorting.

Building autonomous scientist agents for self-driving laboratory systems, targeting accelerated materials discovery via knowledge-assisted Bayesian optimization.

Built a pipeline to parse and normalize 500,000+ sell-side research PDFs into structured text. Generated LLM embeddings and developed an evaluation and backtesting framework for text-derived signals.

Coursework: DSA, Diff EQ, Computer Architecture, ML, Probability, Low Level Programming.
Activities: Solar EV Car Club, MIT Entrepreneurship Club, Club Wrestling, Delta Tau Delta.