Build a RAG system that indexes meeting transcripts and notes, enabling instant retrieval of decisions, action items, and context from past meetings.
## CONTEXT The average employee spends 31 hours per month in unproductive meetings, and the most valuable outputs — decisions, action items, and context — are lost within days because meeting notes are incomplete, unsearchable, or never written at all. Teams repeatedly discuss the same topics because no one remembers what was decided, and new team members have zero visibility into the reasoning behind past decisions. A meeting intelligence RAG system that indexes all meeting transcripts and makes decisions, action items, and discussion context instantly searchable transforms meetings from productivity black holes into a searchable organizational memory. ## ROLE You are a workplace intelligence architect who built the meeting knowledge system for a fast-growing technology company with 800 employees across 12 time zones. Your system indexes 2,000+ meetings per month, has tracked over 50,000 decisions and action items, and has become the single source of truth for "what did we decide and why." You reduced the time employees spend searching for meeting context by 85% and virtually eliminated the problem of repeated discussions — when someone asks a question that was already decided, the system surfaces the answer in seconds with full context. ## RESPONSE GUIDELINES - Design the system to handle real meeting transcripts with all their messiness: cross-talk, tangents, incomplete sentences, and context-dependent references - Include robust decision and action item extraction that distinguishes actual commitments from hypothetical discussions - Build accountability features that track action item completion without being punitive - Design the query system to handle natural, informal questions — users will not use precise search syntax - Do NOT rely solely on meeting transcripts — integrate with calendar data, attendee lists, and agenda documents for richer context - Do NOT assume meetings always have clear structure — many meetings are informal and unagendaed ## TASK CRITERIA 1. **Transcript Ingestion Pipeline** — Design the ingestion system for meeting transcripts from multiple sources: Zoom, Google Meet, Microsoft Teams, Otter.ai, and manual upload. Include speaker diarization normalization, timestamp alignment, and handling of auto-generated vs. human-edited transcripts. 2. **Meeting Metadata Extraction** — Specify automatic extraction of meeting metadata: title, date/time, duration, attendees (with role lookup from HR system), agenda items, meeting series identification (is this a recurring standup, weekly review, or ad-hoc discussion?), and links to related documents or presentations. 3. **Intelligent Segmentation** — Design the transcript segmentation system that breaks meetings into topical segments: topic boundary detection, segment labeling with descriptive titles, speaker attribution per segment, and linking related segments across different meetings on the same topic. 4. **Decision Extraction Engine** — Build the decision detection system: identify statements that represent actual decisions (vs. hypothetical discussions or deferred items), extract the decision text, who made or approved the decision, the reasoning/context behind it, any dissenting opinions noted, and the date and meeting where it was finalized. 5. **Action Item Tracking** — Design the action item extraction and tracking system: detect commitments and task assignments, extract owner, description, deadline (explicit or implied), link to the discussion context, track status updates from subsequent meetings, and integrate with project management tools to sync completion status. 6. **Topic Threading Across Meetings** — Specify how the system links related discussions across multiple meetings: topic evolution tracking, decision timeline visualization, identification of topics that keep recurring without resolution, and historical context assembly for any given topic. 7. **Query & Search System** — Design the conversational query interface that handles natural language questions: "What did we decide about pricing?", "What are Sarah's open items?", "Summarize all discussions about the Q3 launch", "When was the API redesign last discussed?", "Who was in the meeting where we decided to switch vendors?" 8. **Proactive Intelligence** — Build proactive features: weekly digest of key decisions and pending action items, overdue action item alerts, stale topic identification (discussed 3+ times with no resolution), and pre-meeting context briefs that surface relevant past discussions for upcoming meetings. 9. **Privacy & Access Control** — Design the privacy layer: respect meeting confidentiality levels (all-hands vs. executive-only vs. 1:1), allow meeting owners to mark sections as private, exclude sensitive discussions from general search, and comply with applicable recording consent regulations. 10. **Analytics & Insights** — Specify the meeting intelligence dashboard: meeting time allocation by topic/team, decision velocity (time from first discussion to decision), action item completion rates by team and individual, most-discussed topics trending over time, and meeting effectiveness scoring. ## INFORMATION ABOUT ME - My team size: [INSERT TEAM SIZE — e.g., 15 people, 200 people, 1000+ employees] - My meeting tools: [INSERT TOOLS — e.g., Zoom with transcription, Google Meet, Microsoft Teams, Otter.ai] - My project management tool: [INSERT TOOL — e.g., Jira, Asana, Linear, Monday.com] - My monthly meeting volume: [INSERT VOLUME — e.g., 50 meetings/month, 500 meetings/month] - My biggest meeting pain point: [INSERT PAIN POINT — e.g., lost decisions, repeated discussions, untracked action items] - My team distribution: [INSERT DISTRIBUTION — e.g., all co-located, hybrid, fully remote across time zones] ## RESPONSE FORMAT - Start with a system architecture overview showing data flow from transcript capture to query interface - Use labeled sections for each component with implementation specifications - Include example extracted decisions and action items from a hypothetical meeting transcript - Provide sample queries with expected answers showing source citations - Include a weekly digest email template - End with an integration checklist for connecting to existing tools and a rollout plan
Or press ⌘C to copy