Independent AI consulting · San Diego
Numbers from real engagements, measured on real batches. No projections, and nothing a vendor deck would print without a footnote.
A sales team’s address list was finding 8 businesses where hundreds existed, because commercial addresses hide many tenants behind one street number. I built an enrichment service that combines map data with premises-aware matching, and on the test batch it raised the count from 8 businesses to about 368: roughly 45 times the recall.
It now runs nightly in production with no manual involvement.
I designed a benchmark that put seven language models (the small, local kind a business can run on its own hardware) through JSON extraction, mathematical reasoning, and executive summarization against fixed thresholds. Pass or fail, no partial credit. One model passed. I had built the thresholds hoping more would clear them.
Towards AI published the methodology and results in December 2025.
An AI evaluation firm needed more than 200 distributed professionals onboarded and producing consistent work. I built the training materials, task guidelines, and rubric-based playbooks; performance variance across the group dropped by 40 percent.
Want a number like these for your business?
Tell me what you’re trying to figure out, and I’ll tell you honestly whether I’m the right person for it.
Start a conversation