VEQRA is an existing Microsoft 365 automation platform I built that detects and routes enterprise incidents. But it couldn't answer the critical questions: Why did this happen? What is the financial impact? What should we do right now?
The Qwen Cloud Global AI Hackathon was the opportunity to build the intelligence layer VEQRA was missing.
What it does
VEQRA AI orchestrates three specialized AI agents that resolve a critical enterprise incident in 13 seconds:
- Memory Agent — searches historical incident database, identifies similar past cases, determines root cause with confidence score
- BI Agent — calculates financial impact, projects SLA breach risk, assigns criticality score
- Action Agent — generates structured action plan: Teams task, email to Data Owner, Power BI dashboard update
Demo scenario: a critical Leasing VIP contract (€120,000 OVERDUE) is resolved in 13 seconds with zero human intervention.
How I built it
Each agent calls Qwen3-235B directly via Alibaba Cloud DashScope, using the OpenAI-compatible endpoint. Agents communicate through structured JSON outputs, coordinated sequentially by a Python orchestrator.
Challenges
Designing prompts that produce consistent, structured JSON outputs across all three agents was the hardest part — keeping the total resolution time under 15 seconds required careful prompt engineering.
What I learned
Depth beats breadth. One perfect scenario executed flawlessly is more valuable than five agents that partially work. Qwen3-235B's long-context reasoning and structured output capabilities made the multi-agent orchestration reliable and deterministic.
GitHub: https://github.com/nabilfattouch1/VEQRA-AI
Demo: https://youtu.be/zakd6bsdzDA
Top comments (0)