Kevin Street
I build AI and data systems for businesses, and I test them well enough to say whether they're ready for real use.
Plenty of AI demos look great in a conference room and then quietly fail once real people depend on them every day, which is why everything on this site is built so that someone other than me can check it. Each system does one narrow job. One reads an EEOC charge form and fills in a case record, with the quoted evidence sitting next to every field. Another answers questions using only a set of approved documents, and says so when the documents don't cover the question. Every project has a test set and a plain results table, and I'd rather you read that table than take my word for anything.
Projects
Each project is a working application. Its page has the problem, the design, the test set, the results and a changelog. Numbers appear when the evaluation has run; until then the page says so.
| Project | What it does | Evidence |
|---|---|---|
| EEOC charge intake: LLM extraction with source spans and human review | Turns an EEOC charge PDF into a reviewed, structured intake record. Every field cites the page and text it came from, and nothing is exported until a person approves it. | 100% field accuracy, 40 documents, Oct 17, 2026 |
| Grounded Q&A over approved employment-law sources, with abstention | An assistant that answers only from sources someone approved, and cites the passage it used. When the sources don't cover a question, it says so rather than guess. | In progress |
| New-matter intake agent with scoped tools and approval gates | An agent reads an intake email, runs a conflicts search with read-only tools and drafts the matter. It cannot open the matter until an approver signs off. | In progress |
Work with me
If you have an AI or data problem, or a question you've been meaning to ask someone, email [email protected] with a sentence or two about it and we'll set up a time to talk. Follow us on LinkedIn and GitHub.