TSGBuilder
Multi-agent pipeline that writes troubleshooting guides for Azure AI support engineers.
70–80% reduction in guide-authoring time; rolling out across Microsoft support engineering.
Problem
Azure AI support engineers spend hours manually researching and writing troubleshooting guides (TSGs) for new issues. Each guide requires pulling context from multiple internal sources, structuring diagnostic steps, and verifying accuracy, which is a process that doesn't scale with the pace of AI-accelerated product development.
Approach
Built a multi-agent pipeline where specialized agents handle research, writing, review, and PII checking as separate stages with structured quality gates between them. Each agent has a focused role: the researcher gathers context from internal knowledge bases, the writer produces structured TSG content, the reviewer checks accuracy and completeness, and a PII scanner ensures compliance before publishing. The pipeline is orchestrated with GitHub Copilot and runs against Azure AI's internal tooling.
Result
Cut guide authoring time by 70–80%. The pipeline produces consistently structured, reviewed guides that meet compliance requirements. Now rolling out as a hosted service across Microsoft support engineering, beyond the original Azure AI team.