Chat-to-data and knowledge bases
“Nobody can get a number without waiting on the one person who knows the data, and two teams report two different figures for the same thing.”
I help small businesses ask questions of their own data in plain English and get correct answers.
It rests on a knowledge base that records what your data means.
What it is
A way for people to ask questions of company data in plain language and get correct answers, built on a knowledge base that records what the data means.
Who it is for
A business whose people wait on someone else for numbers, or get different numbers from different teams.
Example problems
- “What were sales by region last month?” without waiting for an analyst.
- Numbers that differ between teams because they define a metric differently.
- An AI tool that writes confident, wrong queries.
What you get
- A documented knowledge base of your tables, metric definitions and caveats, kept in a form both people and AI can read.
- A question-answering layer on top.
- A log of every question and query.
How I approach it
- I capture the business’s own definitions first, since that is where AI answers go wrong.
- I keep the human-readable and machine-readable versions in step.
- I log everything so it can be audited.
Why me for this
I built chat-to-data products and knowledge bases that ground AI, at a real-estate software company.
I designed a knowledge system for text-to-SQL agents. It is a completed design and worked sample, not a finished product.
What it can run on
Delivered on Anthropic’s tools, on Google’s tools including Gemini Enterprise and BigQuery, or on open-source models hosted on your own hardware.