Quality review for support chats · Lab

A mock-data app for reviewing customer support conversations against a scoring form, where every question is answered by a rule, by AI, or by a person; built by Barış Ülgen at CogniScope for a prospect demo.

What it is

FIG. 1One ticket through a review. A person signs off every review. The AI pass sorts each question into answered, needs confirmation or human judgement.
Source: autoqa src/types.ts (ResolverType, resolverState), server.ts, docs/superpowers/specs/2026-07-27-home-dashboard-and-integrations-design.md, HEAD d1f7ee5

Support teams grade a sample of their agents' conversations against a scoring form. In AutoQA every question on the form says who answers it: a fixed rule, AI, AI with a person confirming, or a person alone. Each question has a weight, and one answer can be marked as a critical failure, which sets the score to zero. You can describe the form you want in a sentence and the app drafts it.

An evaluation runs an AI pass over a batch of tickets first. Each question comes back answered, waiting for a person to confirm, or left for human judgement. The reviewer then opens each conversation, checks or overrides the answers, fills in the rest and publishes the review, with a button that jumps to the next ticket. A home screen lists the pending reviews and score trends.

In the demo, the tickets, the trends and the AI pass are all mock data. The one live AI call drafts forms.

What happened

Built by Barış Ülgen at CogniScope, over about ten days in the summer of 2026, for a demo to a prospect, and shown in that meeting. Nothing after the demo is on record.