Case study
Meet Howard.
Texas school finance runs on thousand-page rulebooks — FASRG, EDGAR, the federal Uniform Guidance. The people who have to follow them needed a faster way to ask "how?" So we built them an expert.
- Product
- Howard — schoolbusinessbot.com
- Scope
- AI assistant: design, document pipeline, accounts, operations
- Stack
- Python · FastAPI · retrieval-augmented LLM · self-hosted
- Audience
- School business professionals
The problem: the answer exists, somewhere
Ask a school business office a question — "Is this an allowable cost?" "What function code does this belong to?" — and the answer usually lives in one of two places: a regulation nobody has time to re-read, or the head of the one person who's been doing this for twenty years. New business managers get neither.
Howard is an AI assistant trained on the documents themselves: the Texas Financial Accountability System Resource Guide (FASRG v19), EDGAR, and 2 CFR 200. Ask in plain English; Howard answers from the source material and points you back to it — because in this field, "trust me" isn't a citation.
The part you don't see
A chat box is the easy bit. The real scope is everything underneath: a document pipeline that turns dense regulatory PDFs into clean, searchable, citable passages; retrieval tuned so answers come from the right section of the right document; a test suite of real questions that every change must pass before it ships; and the unglamorous rest — user accounts, email verification, private-question mode, question history, and an admin view.
There's even a feedback loop: a public "see what people are asking" feed, and the questions users actually ask shape the training material we build next. The tool gets better because people use it.
Howard is live, self-hosted, and answering real questions from real school business professionals — free to use with a sign-in at schoolbusinessbot.com. Well done, Howie.
Start your project