TestForge AI

AI-powered Playwright tests with self-healing locators

4 followers

AI-powered Playwright tests with self-healing locators

4 followers

QA teams spend half their week on plumbing — boilerplate, brittle locators, flaky triage. TestForge AI removes it. Paste a requirement → TestForge drafts Gherkin, derives the POM via Microsoft Playwright MCP, scaffolds Playwright TypeScript, runs in disposable containers. Claude classifies failures (real bug vs flake) and explains in plain English. Backed by 6 peer-reviewed papers (IEEE, Wiley, Elsevier, Springer). MIT-licensed npm suite at ~5,400 weekly downloads.

Launch tags:Productivity•SaaS•Artificial Intelligence

Launch Team

Render: Get up to $100K in startup creditsThe fastest way to launch and scale your startup in the cloud

Promoted

Hunter

📌

Hey Product Hunt 👋 I'm Vijay — independent researcher and the sole developer behind TestForge AI. The 30-second pitch: QA teams spend half their week on plumbing. Writing boilerplate, fixing brittle locators, triaging flaky failures. TestForge AI removes the plumbing. You paste in a requirement document. TestForge drafts a complete suite of Gherkin scenarios for review. Approved scenarios become real, executable Playwright tests automatically — TestForge derives the page-object model from your live application by scraping the actual DOM via Microsoft Playwright MCP, picks the right selectors, and scaffolds test files across Chromium and Firefox. You don't write the Playwright code, you don't write the page objects, you don't write the assertions. Every regression run spins up a clean disposable browser container, captures pixel-by-pixel screenshots, and produces a structured report. When something fails, an AI analyst built on Anthropic's Claude classifies the failure (real bug vs flake), explains it in plain English (Category / Expected / Actual / Likely cause / Suggested fix), and drafts the Jira ticket if escalation is warranted. Self-healing locators recover from common UI changes — renamed buttons, shifted CSS classes, restructured sections — using role, label, accessible name, and adjacent-element text. Over months, the system learns your codebase's failure patterns. Behind the platform is real research: six journal manuscripts under peer review at IEEE Access, IEEE Software, Wiley STVR, Elsevier JSS, and Springer EMSE. Three sole-authored MIT-licensed npm packages (@vijaypjavvadi/bdd2pw, sel2pw, pw-emit) with combined ~5,400 weekly downloads. Datasets permanently archived on Zenodo. Honest disclosure: TestForge AI is in Beta. We're early. If you try it and hit something rough, please reply here — I read every comment and ship fixes fast. Try it: https://testforge-ai.com GitHub: https://github.com/javvadivijayp... Research: https://vijayjavvadiresearch.ai G2 listing: https://www.g2.com/products/test... Ask me anything 🙏

Report

5mo ago

Forum Threads

p/testforge-ai

•

5mo ago

What's the worst time-sink in your test automation workflow right now?

Building TestForge AI in public launching here on PH next Tuesday (June 9) and I want to gut-check the problem framing with this community before launch day.

My working hypothesis: most QA teams spend half their week on plumbing writing Page Object boilerplate, fixing brittle locators after every UI change, and triaging flaky failures that turn out to be environment issues, not real bugs. The actual signal (real defects) gets drowned in the noise.

So I built TestForge AI to remove the plumbing: paste a requirement it drafts Gherkin scenarios generates the Playwright TypeScript test files (using Microsoft Playwright MCP to scrape the live DOM and pick stable selectors) runs everything in disposable containers when something fails, an AI analyst built on Anthropic's Claude classifies it (real bug vs flake) and explains it in plain English.

The technical bet: deterministic-first, AI-second. Rules engine handles the common cases instantly; Claude only gets consulted for the 1-2% of edge cases where the deterministic layer is uncertain. Every classification shows you why it was made.

View all