Veta: AI agent that QA-tests Android apps (github.com)

Veta, an innovative AI-powered QA testing agent for Android apps, has been introduced as a solution for visual, functional, and accessibility testing. Designed for the AMD Developer Hackathon, this autonomous swarm technology leverages AMD GPUs exclusively for all AI inference tasks, running 100% on AMD hardware without fallback options. Veta replaces traditional scripted UI tests by enabling an AI model to interpret plain-English task descriptions, interact with the Android interface using ADB commands, and produce detailed reports mimicking human QA testers—complete with pass/fail verdicts and session analysis. The significance of Veta lies in its seamless integration of AI with mobile app testing, employing a unique architecture that includes planning, execution, verification, and reporting stages. Each testing session can run concurrently across up to 100 containerized Android instances, enhancing throughput and efficiency. The agent employs advanced features like ground-truth screen tracking and automated loop detection, ensuring more reliable test outcomes. By normalizing testing procedures that typically require significant human input, Veta not only accelerates the testing cycle but also empowers developers to deliver higher-quality applications faster, ultimately contributing to the evolution of QA practices in the AI/ML landscape.