Ideation with an AI critic
Requirements start as a conversation, not a form: the interviewer asks, a deterministic critic scores coverage, and only a spec that clears the bar reaches a human for approval.
How it works
The interviewer asks in context — your brief, a repo-scan overview, and your component library all inform what it asks next.
No LLM judgment call here: a depth-weighted rubric decides whether an answer area is covered, thin, or missing — consistently, every time.
Thin or skipped answers are tracked, and a feedback loop steers future questions away from framings that keep producing low-signal answers.
A completed interview becomes a four-part requirements artifact in one shot, with an automatic quality check and one repair attempt before it goes to a human.
The requirements gate is where a person, not the critic, decides the spec is ready to build from.
The agents behind it
Human control
Requirements gate — a human approves the synthesised spec; the critic only decides when the interview itself is deep enough to write one from.