# TestForge-AI: Autonomous Software Testing and Failure Triage Powered by AMD ROCm and Lemonade **URL:** \ **Category:** Showcase\ **Tags:** ai, rocm, ryzen-ai, gpu, developer-program\ **Created:** [September 1, 2026, 12:42am UTC](https://devcommunity.amd.com/t/testforge-ai-autonomous-software-testing-and-failure-triage-powered-by-amd-rocm-and-lemonade/937 "2026-09-01T00:42:58Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![Ganesh\_Yamavarapu](https://avatars.discourse-cdn.com/v4/letter/g/e495f1/32.png) [@Ganesh\_Yamavarapu](https://devcommunity.amd.com/u/Ganesh_Yamavarapu)\ **Post date:** [September 1, 2026, 12:42am UTC](https://devcommunity.amd.com/t/testforge-ai-autonomous-software-testing-and-failure-triage-powered-by-amd-rocm-and-lemonade/937/1 "2026-09-01T00:42:58Z")
Excited to introduce **TestForge-AI** , an open-source, local-first software-testing agent designed to automate test-failure analysis and defect triage using **AMD ROCm** and the **AMD Lemonade SDK**. Modern automated tests generate large amounts of evidence, including stack traces, screenshots, browser-console errors, failed network requests, retry histories, and accessibility violations. QA engineers often spend considerable time correlating these artifacts before identifying a likely root cause. TestForge-AI brings this information together and transforms it into a structured, evidence-based triage report. ### Core Features - Parses failed tests from Playwright JSON reports. - Collects error messages, execution duration, browser details, retry history, console errors, and network failures. - Reviews accessibility findings and incorporates their impact into defect severity. - Uses PyTorch with AMD ROCm to calculate visual-regression metrics from screenshots. - Applies deterministic severity rules so defect priority remains consistent and auditable. - Sends the grounded evidence package to a locally hosted AI model through the AMD Lemonade OpenAI-compatible API. - Produces a structured report containing the likely root cause, confidence level, supporting evidence, recommended actions, regression-test suggestions, and human-review requirements. - Provides an evidence-only mode that can operate in CI/CD pipelines without language-model inference. ### Why AMD ROCm and Lemonade? AMD ROCm enables GPU-accelerated screenshot comparison and visual-regression analysis. The AMD Lemonade SDK provides local AI inference, allowing sensitive test artifacts to remain on the developer’s computer. Application screenshots, internal logs, API failures, and test data do not need to be transmitted to an external cloud service. ### Responsible AI Design TestForge-AI separates measurement from AI reasoning. Deterministic tools collect evidence and calculate severity, while the local language model explains the available findings. The model is instructed not to invent logs, selectors, source-code locations, or network requests. It also cannot execute generated code or automatically modify the application repository. Low-confidence conclusions are marked for human review. ### Example Usage ```auto testforge doctor testforge analyze playwright-report.json \ --baseline-dir baselines \ --output testforge-report.json ``` The project currently supports Playwright reports, with planned support for Selenium, Cypress, JUnit XML, REST API testing, failure clustering, duplicate-defect detection, and CI/CD integrations. I welcome feedback from the AMD developer community regarding additional testing frameworks, local models, performance optimizations, and AMD-accelerated capabilities that could improve future versions. **GitHub repository:** [TestForge-AI](https://github.com/ganeshyamavarapu1018/TestForge-AI) **Demo video:** Coming soon The project is distributed under the **Apache License 2.0**. ---
**Author:** ![Ganesh\_Yamavarapu](https://avatars.discourse-cdn.com/v4/letter/g/e495f1/32.png) [@Ganesh\_Yamavarapu](https://devcommunity.amd.com/u/Ganesh_Yamavarapu)\ **Post date:** [September 1, 2026, 1:07am UTC](https://devcommunity.amd.com/t/testforge-ai-autonomous-software-testing-and-failure-triage-powered-by-amd-rocm-and-lemonade/937/2 "2026-09-01T01:07:52Z")
![TestForge-AI](https://global.discourse-cdn.com/amd/original/2X/f/ffd4abf37e757e35fd2a7089644661a39173d8f1.jpeg) ---
**Author:** ![Hakob\_Arzumanyan](https://dub1.discourse-cdn.com/amd/user_avatar/devcommunity.amd.com/hakob_arzumanyan/32/214_2.png) [@Hakob\_Arzumanyan](https://devcommunity.amd.com/u/Hakob_Arzumanyan)\ **Post date:** [September 1, 2026, 12:56pm UTC](https://devcommunity.amd.com/t/testforge-ai-autonomous-software-testing-and-failure-triage-powered-by-amd-rocm-and-lemonade/937/3 "2026-09-01T12:56:11Z")
Hello, thanks alot for sharing, one idea on where the GPU could actually earn its keep. Right now ROCm is doing pixel diffing, which is about the cheapest job you could hand a GPU. A screenshot diff is a few megabytes of subtraction, and shipping it to the card may cost more in transfer time than it saves. Your roadmap already has VLM screenshot interpretation through Lemonade, and that’s the job that genuinely needs a GPU. It’d also give you something a pixel diff can’t, like telling a real layout break apart from a font rendering difference. I’d move it up the list.