Hi, I’m Dave
I build trustworthy AI production systems.
My focus is on the gap between “it works in the demo” and “we can hand this to an auditor.” That gap is where most AI data pipelines live, and closing it is the work I find most interesting.
What I build
- Trust-retrofit for AI extraction pipelines — labeled eval sets, accuracy and failure reports, confidence scoring, audit logs, PII redaction. The layer that makes an existing pipeline something you can actually stand behind.
- Document processing pipelines — structured extraction from invoices, contracts, receipts, statements, and whatever other documents your operation runs on. Built in Python, not low-code wrappers.
- AI benchmarking and model evaluation — does GPT-4o or a specialized OCR API actually fit your use case? I’ll answer that with numbers, not opinions.
- Secure, self-hostable extraction kits — designed for environments where you can’t send sensitive documents to a third-party API.
The common thread: I care about what happens after the demo. Reliability, auditability, and security are the defaults — not the add-ons you bolt on when something breaks.
Get in touch
If your AI data pipeline needs a reliability layer — or you want to understand what it’s actually doing before something goes wrong — reach out.