Greenwich, CT | 203-564-6079 | kawgit56@gmail.com
Built and scaled distributed, AI-powered pipelines to process SEC filings into structured, searchable data using LLM-driven NLP parsing. Developed end-to-end automated tests.
Optimized the world's 2nd strongest chess engine (Torch) by designing sparse neural network inference improvements, boosting speed by 20% and strongly cutting scaling costs. Built core C++ modules for AI Chess Coach.
Developed a custom transformer-based language model from scratch with PyTorch, trained on Harry Potter and Fineweb-Edu. Features a BPE tokenizer with a novel deduplication scheme, new architectures including GQA sliding window attention, and Muon-inspired optimizer algorithm changes that cut training time by half.
Lightweight chess engine utilizing bitboards, alpha-beta pruning, and selective search heuristics in C++, along with a JS web interface.
Implemented a computer vision pipeline to reconstruct 3D point clouds from video footage.
Built a 3D graphics renderer from scratch by rasterizing triangles directly to the screen utilizing low-level math.
Education: Boston University (BA Computer Science, Expected 2027), Case Western Reserve University.
Skills: Python, TypeScript, C++, Java, SQL, Svelte, Node.js, Bash, PyTorch, NumPy, Matplotlib, OpenCV, Linux, Git, Docker. Focus in ML, Computer Vision, Distributed Systems, NLP, and GPU Acceleration (CUDA).