💬
User Query via Copilot Chat
Natural language question spanning work items, email, documents
Raw user intent
🧠 Intelligence Layer
🎯
Query Analyzer
Intent classification — determines which data sources to query and how to scope each request. Avoids broad, noisy RAG retrieval.
Scoped queries dispatched per source
🔧
Azure DevOps
ADO REST API
🔒 Scoped: Project + Work Item Type
Returns: Work items, repos, pipelines
📧
Microsoft Graph
Graph API
🔒 Scoped: User + Date Range
Returns: Email, calendar, people, files
📄
SharePoint
Search API
🔒 Scoped: Site + Library
Returns: Documents, lists, pages
⚙️ Response Assembly
Step 1
Merge Results
Deduplicate and normalize results across all three sources
Step 2
Relevance Ranking
Score and filter by original query intent
Step 3
Synthesize Response
Azure OpenAI generates grounded answer from ranked results
Grounded Response with Source Citations
Each claim linked to its source document, email, or work item
🔑 Key Insight
Scoped search beats broad retrieval. By analyzing intent first and querying each source with targeted filters, you eliminate the noise that makes RAG answers unreliable.