Plain-language questions, trustworthy answers
A natural-language-to-SQL agent that lets non-technical users query data in plain language — designed so they can trust and verify every answer.
The challenge
Non-technical users had to open dashboards or wait on analysts to answer everyday data questions. Standard and custom reporting couldn't keep up with the long tail of questions people actually had — and in a regulated, enterprise context, any self-serve answer still has to be trustworthy.
What I did
I led development of a natural-language-to-SQL agent: ask a question in plain language, get a chart or answer back. Crucially, I designed it for trust — the query it runs is transparent and auditable, so users can verify the result rather than take it on faith.
Anyone can ask. Everyone can verify.
The outcome
It's in production across 60+ markets, expanding data access well beyond standard and custom reporting — without sacrificing the auditability enterprise customers require. The capability has supported millions in revenue and client retention for some of our largest clients.
Guard Dog — data quality, watching quietly
A background agent I built for continuous, proactive anomaly detection. It surfaces data-quality and pipeline issues — with context and suggested next steps — before they reach downstream outputs. No user action required; it just runs.
What shaped this design
I once shut down a working AI tool my own team relied on — it was improving outcomes, but some PMs had stopped verifying its output and started trusting it blind. That lesson — AI people can't or won't verify becomes a liability — is exactly why this product makes every query transparent and auditable. The full story is on the homepage →
What it proves
Applied LLM product work, accessibility for non-technical users, and responsible-AI design where transparency is a feature, not an afterthought.