DevTeam AI: A Multi-Agent Engineering Team on QwenCloud

Built for the QwenCloud Global AI Hackathon's Agent Society track, DevTeam AI is a multi-agent system designed to mimic a real software engineering organization rather than flattening all roles into a single AI prompt. An orchestrator coordinates 11 specialized agents—covering product management, architecture, backend and frontend engineering, QA, CTO-style review, negotiation, revision coordination, code generation, and code review—each with defined responsibilities, dependencies on prior agents' outputs, and the ability to challenge or negotiate conflicting decisions.

A key design choice is that no stage advances automatically: every step in the pipeline, from idea to PRD, architecture, and final code review, requires explicit human approval, with options to approve, regenerate, or request revisions. This keeps the system closer to a genuine engineering workflow than a simple chained-prompt demo.

The author chose QwenCloud for its OpenAI-compatible API, consistent structured outputs (Markdown, JSON, Mermaid diagrams), and strong reasoning performance on review-heavy tasks like spotting architectural inconsistencies. The stack pairs a Next.js frontend with a FastAPI backend and SQLite storage, using Server-Sent Events instead of WebSockets for simpler, more reliable real-time updates. The project offers a concrete example of how multi-agent AI systems can model complex, approval-driven real-world engineering workflows.