🤓 Latest Submissions
Testloom
AI-generated software can look correct without actually satisfying the behavior it was supposed to implement. Testloom addresses this gap by turning a plain-English feature request into executable browser acceptance checks and then testing the behavior against a live application. A user starts with a simple request such as: “Keep users logged in after refreshing the page.” Testloom uses an LLM to turn that request into structured acceptance checks. The generated plan can be reviewed before execution. Testloom then runs the approved checks in a real Playwright browser against the selected live application, records what happened, captures browser evidence, and produces a deterministic PASS, FAIL, or INCONCLUSIVE result. The demo uses two controlled versions of the same application. In the working version, authentication is stored in a cookie and survives a browser refresh. In the intentionally broken version, the session is stored only in memory and is lost after refresh. This creates a simple demonstration of the core problem Testloom is designed to solve: Working implementation → browser verification → PASS Broken implementation → browser verification → FAIL The target users are developers, hackathon builders, teams using AI coding tools, and anyone who needs to verify that implemented software actually satisfies a requested behavior. Testloom is designed around a simple principle: the LLM generates the test plan, but the LLM does not decide whether the software passed. The final result comes from real browser execution and observed application behavior. The system is built with Next.js, TypeScript, Playwright, Zod, Groq, Vercel, and Vercel Blob.
27 Sep 2026
