loading . . . GitHub - lukemurraynz/NimbusIQ: Azure infrastructure governance and optimization insights as part of the DevDays Hackahon 2026. | Luke Murray A few months ago, I participated in the Microsoft AI Dev Days Hackathon - and I’m excited to share that my submission, NimbusIQ, won the category for Best Azure Integration! 🎉
The hackathon challenged people worldwide to build AI-powered solutions using technologies such as Microsoft Foundry, Microsoft Agent Framework, Azure MCP, and GitHub Copilot Agent Mode, with solutions deployed to Azure and developed through GitHub.
For NimbusIQ, I built a proof of technology for an AI-powered cloud intelligence platform that continuously discovers, scores, and evolves Azure Service Groups.
It also gave me a chance to build beyond the “agent demo” layer: a React and Fluent UI frontend, an ASP.NET Core control plane API, a .NET agent orchestrator, PostgreSQL-backed state, OpenTelemetry instrumentation, managed identity, Key Vault, Azure Container Apps, and Azure Developer CLI deployment.
At its core, NimbusIQ uses a ten-agent analysis pipeline powered by Microsoft Agent Framework and Microsoft Foundry to:
✅ Detect configuration drift
✅ Surface governance-approved remediation plans
✅ Generate deployable Infrastructure as Code
✅ Keep humans in control before any change is applied
It was a great opportunity to spend a few focused days building, testing, and learning more about multi-agent orchestration, agent-to-agent communication, OpenTelemetry integration, and secure connectivity patterns such as Network Security Perimeter.
I also probably spent far too much time trying to make the UI feel as close to the native Azure portal as possible - but that was part of the fun!
I love hackathons because they create space to learn quickly, experiment with new ideas, and be inspired by what others are building. Congratulations to the other award winners and to everyone who participated - there was some amazing work submitted, and some of it has genuinely world-changing potential.
One of the biggest learnings was that the hard part is not just getting agents to generate useful output - it is designing the guardrails around them.
For NimbusIQ, that meant using deterministic checks where consistency mattered, grounding LLM agents in structured Azure evidence, separating analysis from action, tracing agent-to-agent decisions, and requiring human approval before any generated remediation could be applied.
It was also my first real hands-on opportunity to explore hosted agents in Microsoft Foundry Agent Service, and to think through an important architectural decision: which agents should be platform-managed, and which should stay closer to the application for orchestration, debugging, control, and integration - especially when separating deterministic rule-based agents from more agentic, LLM-driven ones.
You can check out NimbusIQ here:
https://lnkd.in/eq7dH7Th
#aidevdays #hackathon #reactor #mvpbuzz #azure #microsoftfoundry #microsoftagentframework #githubcopilot https://www.linkedin.com/posts/ljmurray_github-lukemurraynznimbusiq-azure-infrastructure-share-7459719960806748160-p1ot?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAqDwoABeVWv4tHcb5O0L-FGMqVN6TX5Okk