Public-sector software - 2026
Years of undocumented logic, turned into a searchable knowledge base
Our client's engineers depended on a few people who remembered how things worked. We built them a private knowledge base that answers questions from their own code, inside the tools they already use.
- Scripts documented
- Thousands
- Cost per script
- Under 1c
- Places it lives
- 3
The problem
Thousands of custom scripts had accumulated behind software that public agencies rely on daily, handling rules, calculations, workflows and notifications. Almost none of it was written down. Answering a question meant searching folders by filename, reading code line by line, or finding the one colleague who had solved something similar years earlier. New engineers took a long time to become useful, and the knowledge that mattered most sat in a few people's heads. It all existed; nobody could get at it.
What we built
We started from the question engineers actually ask, "has anyone already solved this?", rather than from the model. Every script was given a plain-language summary of what it does, when it runs and what it touches, and made searchable by meaning rather than by filename. A grounded assistant answers from the client's own code and shows what it used, scoped so one customer's rules never appear in another's answer. It reaches people in the three places they already work: a web app, the company chat workspace and the code editor.
Results
Thousands of scripts are documented and searchable where very little was before. Questions that used to mean waiting on a senior engineer now return a sourced answer in seconds, and new engineers explore years of history on their own from their first week. Everything runs inside the client's own cloud with single sign-on, permissions enforced on every action, and the AI under their existing enterprise agreement, so the data stays under their control.
Let's build together
Let's find out if AI is worth it for you
A short conversation is usually enough to tell whether there is a real case here, and we will say so if there is not.