Built by Mark in Conley, Georgia. I take messy, contradictory data — chemical, municipal, commercial — and build the tool that answers the one question the people who own it can't answer today. Every result traces back to its source, and the same inputs return the same result on someone else’s machine.
Built from 44,648 peer-reviewed papers. Maps 5,240 distinct molecules across 852 ingredients into 296,732 computed relationships, indexed against a 612,557-recipe corpus, with UMAP cross-category bridges surfacing 450 standout ingredients.
$ open compkitchen.com ↗Multi-source signal detection on a FastAPI backend with scheduled ingestion, deduplication and ranking. Drafting is assisted; an editor review queue is the gate, so nothing publishes without a human approving it.
$ open Licensing inquiries ↗Natural-language brief to a working formula: character-impact compounds, OAV-ranked, starting ratios in ppm, mapped to real vendor SKUs, with an iteration rubric. Lets small food brands reason at the molecular level without a flavor house.
$ open Request access ↗Turns one recording into the same voice speaking every language. Transcribes to timed segments, translates each with Claude, re-speaks with real cross-provider failover, times it back to the source, and stores every result with a hash-verified provenance manifest.
$ open cast.rowset.co ↗ $ read the source ↗Reads skin, hair, and eye color from one selfie, computes undertone, depth, and contrast in CIELab, and returns a season, a palette, and the exact reasoning behind it. Digital draping puts a best color and a draining one side by side on your own photo.
$ open truehue.rowset.co ↗ $ read the source ↗Mark builds analysis and data systems where the answer has to be defensible, not just produced. Nine years in K-12 school-district IT operations behind the engineering, and a published correction on his own work when he got something wrong. One principal: the person you brief is the person who writes the code.