
SwarmAudit is a multi-agent audit workspace that helps developers understand whether a repository is ready for real-world use. Instead of acting like a single chatbot reviewer, SwarmAudit coordinates specialized agents for security, performance, code quality, documentation, dependency risk, error handling, observability, configuration safety, and CUDA-to-ROCm portability. Users enter a public GitHub repository URL, and the app clones the project, analyzes the source files, ranks findings by severity, and presents a clear remediation-focused report. The project is designed for teams building with fast-moving modern codebases, including AI-assisted development workflows where code can appear functional but still miss important production safeguards. SwarmAudit turns those risks into an actionable dashboard with severity filters, file references, explanations, suggested fixes, production-readiness scoring, category summaries, and Markdown/JSON exports. For the AMD Developer Hackathon, SwarmAudit integrates Qwen2.5-Coder through an OpenAI-compatible vLLM backend running on AMD Developer Cloud with ROCm. This allows the scanner to enrich static findings with more contextual code explanations when GPU access is enabled. The public Hugging Face Space remains available as a reliable demo, while the AMD/vLLM path demonstrates how the system can scale into a richer AI-agent review workflow powered by open-source models and AMD infrastructure.
10 May 2026