I'm a software engineer who loves making data go fast (safely!). Interested in database internals, ML concepts, and optimizing/evaluating agentic systems.
- A home-cooked recurrent NN for predicting ETF prices
- A home-cooked visual question answering system, scoring as high (66.76) on VQAv2 as models 10x its size!
- Contributions to column
- Published ML research on improving JPEG image quality as a research assistant
- A Google Cloud Function for Twitch chat to trigger a streamer's smart lights using webhooks (built live on stream!)
My home-cooked VQA model inspired a litany of other ML project ideas, which I've made the decision to store under one GitHub organization, Cothogonal due to them being tightly-coupled subtrees from a main, private repo, allowing me to hide all my personal junk docs and only publish clean, readable source repos.
- SGOCR is an open-source, spatially-grounded, OCR-focused dataset generation pipeline for turning images with text in them (documents or natural scenes). A 6k sample has been published on Hugging Face for free use.



