📥 Stage 1 — Meeting Capture
Source meeting with recording and transcript enabled.
🔗
Custom Graph Connector
REUSABLEDecodes base64 chat ID → join URL → meeting ID → transcript ID → raw WebVTT.
🟣 Stage 2 — Scribe Agent
Raw WebVTT fetched via Graph API. Saved to archive first.
AI extracts key decisions, action items, open questions, follow-ups.
Word doc generated via Word Online connector.
Documents saved to SharePoint
📁 Stage 3 — Two SharePoint Libraries
🗄️
Library A — Raw Transcripts
IMMUTABLEOriginal transcripts. Read-only archive — never modified after save.
📋
Library B — Generated Summaries
EDITABLEAI-generated summaries. Reviewable, approvable, editable.
💡 Two libraries, not one — clean separation prevents accidental modification of originals and creates a clear audit trail.
Summary routed for review
⚡ Stage 4 — Approval Gate
Reviews summary against raw transcript for accuracy.
✅
Approve / Request Corrections
Approved docs → knowledge base
🟢 Stage 5 — Library Agent
Copies approved docs to Azure Blob Storage for indexing.
Indexes content for full-text and semantic search.
Natural-language questions across all meeting knowledge.
🚀 Built in 2.5 days at a customer hackathon — "What if every meeting your team runs was automatically documented, approved, and searchable by end of day?"