AI governance tool, 2026

Narrative Guard

Audits a draft against a team's own brand rules before it goes out, and says where it breaks them.

Problem: Brand guidelines are a document nobody rereads. By the time a post, an email or a press release is off-message it has usually already been approved by someone going from memory.
What we built: A team writes its brand constitution once and it is indexed into a RAG namespace, so an audit is scored against the rules that were actually retrieved rather than against whatever the model believes about the brand. The model writes the report and the findings; the score itself is computed in the backend, which is what stops two runs over the same draft from disagreeing. Workspaces carry owner, admin and member roles, and a finished report exports as a PDF.
My part: I built the analytics page: the charts, the filters and the Convex queries behind them.
Result: Deployed and usable, covering social posts, emails, ads, press releases and website copy, with trend analytics across saved reports.
Stack: Next.js, TypeScript, Convex, RAG, Gemini API