From a user story to AI-assisted tests, with review at each stage.
01 · Context
From a user story to AI-assisted tests, with review at each stage.
The problem
AI can generate tests quickly. Someone still needs to check whether they cover the requirement, make the right assertions, and produce useful results.
My role
I built this personal project to connect AI-assisted test creation with executable checks and human review.
02 · Engineering decisions
How it works.
- 01Keep four explicit human review gates in the workflow.
- 02Validate artifact schemas before moving between stages.
- 03Connect acceptance criteria, risk, tests, and execution evidence.
- 04Constrain automated healing and preserve human responsibility for approval.
- Story + risk
- Human review
- Playwright · Newman
- Evidence + report
03 · Quality strategy
How I tested it.
Playwright E2E and Postman/Newman API checks sit alongside schema validation and guarded repair behavior. Example runs and pilot evidence are available in the repository.
04 · Result & reflection
Results so far.
The workflow links requirements to tests and their results. A small pilot records the benefits alongside the time needed for review.
Scope & limitations
Qaizen helps create and run tests. Its expected-dataset results don’t establish how reliably a live model generates correct tests.
What I learned
Each review step needs to show the reviewer what to check and what they’re approving. Otherwise, it adds a pause without helping them assess the test.
