Inspiration
What it does
How I built it
Challenges I ran into
Accomplishments that I'm proud of
What I learned
What's next for test-engineer-agent
Inspiration
Requirements rarely arrive in a form that is ready to test. Test Engineer Agent makes the path from a document or work item to defensible test coverage deliberate: parse requirements, add product context, review the plan, generate cases, and only then export or run approved candidates.
What it does
Test Engineer Agent accepts Markdown, Word, and Excel requirements, and can import and sync work items from Jira Cloud and Azure DevOps Services. It extracts requirement items, enriches them with grounded UI, API, and workflow facts from linked product materials, and creates reviewable use cases and test cases with coverage diagnostics and traceability.
QA projects retain staged snapshots and show blockers, next actions, runs, and reports. After human review, users can export suites as CSV, Excel, or JSON. The Automation workflow previews which cases are executable, manual, unsupported, or invalid; approved executable candidates can run through Playwright and return result and report evidence.
How we built it
The app combines a React + Vite frontend with a Python/FastAPI backend and Pydantic contracts. Gemini through Google ADK/GenAI supports the requirement and test-generation workflows. Firebase Authentication and Cloud Firestore support identity and durable workflow data. Jira Cloud and Azure DevOps adapters connect source work items, while an isolated Node.js Playwright runtime executes approved browser candidates. Docker and Google Cloud Run support deployment.
Challenges we ran into
The central challenge was keeping generated work grounded and traceable instead of treating model output as final. We separated machine quality signals from durable human approvals, preserved immutable workflow snapshots as upstream changes made downstream artifacts stale, and classified unsafe, manual, and unsupported automation before execution. We also built diagnostics and deterministic fallbacks for malformed model output and unavailable integrations.
What we learned
AI assistance is most useful when every output has provenance, coverage and traceability evidence, visible diagnostics, and a human decision point. Automation needs its own readiness gate: not every generated test case should become executable code.
What's next
We plan to productize the experimental Playwright page-object-model generation path, expand verified integrations and real-app execution grounding, and continue strengthening reports and review workflows.
Log in or sign up for Devpost to join the conversation.