One founder,
a research machine.

Shovel is one founder running an AI research machine that does custom research and data acquisition. The mission: take data that the source agencies publish but don't make easy to use, and turn it into a clean, queryable, analyzed dataset. Sometimes that's a public product; sometimes it's built on assignment for an organization that needs data nobody has organized yet. MSHA mining incidents back to 1983. OSHA workplace-safety records across every industry. COMEX and LME precious-metals ticks every second. I set the playbooks and check the output; automated scrapers and LLM-assisted pipelines do the volume.

The aesthetic is Pieter Levels meets Bloomberg Terminal: small, sharp, opinionated. Every product in the portfolio started as a question someone should have been able to answer with public data but couldn't, because nobody had organized it.

The pipeline is the product. Automated CI, self-healing scrapers, LLM-assisted triage, and a meta-brain that watches the whole stack and files its own bug reports. I design and review; the machine executes 24/7, and the data compounds.

Nothing here is true because a model said so. Every number traces back to a source document, the sources are published next to the finding, and anything that looks unusual gets read by hand before it goes out. The AI supplies the reading volume; the judgment call is mine, and my name is on it.

We also build the software on top of the data. When the deliverable is a working tool rather than a raw dataset (a search UI, a review pipeline, an internal dashboard), the same machine designs it, ships it, and keeps it running.

The same pattern extends to distribution. The pipelines follow each product's niche on X, read the conversations, and draft replies and posts in the house voice — but a human approves before anything ships. They advise the accounts; they don't run them.