§About
We build document retrieval systems for organizations that work with material they can’t hand to a third party. The person who scopes your work is the person who writes it, and you deal with them directly from the first conversation onward.
PhD in electrical engineering, with research in machine learning. Built the verified-citation retrieval engine this practice is based on: the ingestion and chunking, the authorization model, the constrained generation, and the checking pipeline that decides whether an answer gets shown at all.
Writes and teaches on the subject, including an interactive lesson on what agentic AI actually means and a YouTube channel working through the engineering behind it.
Why this focus
The demo is easy. What is hard is the version that survives someone checking it: where a citation cannot be invented, a quotation cannot drift, a restricted document cannot be retrieved by someone who should not see it, and the system says so when the documents disagree with each other.
That work is unglamorous and it is most of the value. It is also the part that decides whether a system gets used after the first month, or quietly abandoned because nobody trusts what it says.
That is the work this practice takes on, most often for people whose documents are sensitive enough that the answers have to be checkable.
We will show you the system running and give you a straight answer about whether it fits what you are trying to do.
