SignalBrief — AI Research Intelligence Assistant

SignalBrief is a source-bounded research intelligence assistant. It analyzes only the material supplied by the user and separates:

Source-supported claims

Inference

Contradictions and evidence gaps

Unresolved unknowns

Key findings and open questions

The system is designed to show what the evidence actually supports, and what it does not.

The live system uses a GitHub Pages PWA, a private Cloudflare Worker backend, and the OpenAI API. API credentials remain server-side rather than being exposed in the client.

I defined the workflow, source boundaries, evaluation rules, failure handling, UX behavior, and acceptance criteria, with AI tools assisting implementation and iteration.

SignalBrief turns a common AI weakness into an explicit design constraint: the model is not allowed to quietly fill gaps with outside knowledge.

The result is a research workflow that makes uncertainty inspectable instead of merely producing a polished answer.

SignalBrief is deployed and available as a working PWA.