AI Meeting Intelligence Platform
Project Overview
We designed and developed a full-stack AI meeting intelligence platform that automates the entire post-meeting workflow. The platform integrates with users' calendars and meeting tools, deploys AI agents to join scheduled meetings, captures and transcribes conversations with speaker diarization, and uses LLMs to generate structured meeting outputs. A key technical implementation was the use of Model Context Protocol (MCP) to connect the AI layer directly with Jira, Azure Boards, and Trello — enabling the platform to create tasks, update boards, and sync action items through standardised AI tool calls rather than fragile custom API wrappers. A searchable meeting knowledge base built on vector embeddings allows teams to query past discussions, decisions, and commitments at any time.
The Challenge
- Teams were spending hours after every meeting manually writing notes, preparing minutes of meeting, identifying action items, and updating project management tools — a significant and recurring productivity drain.
- Action items and commitments discussed in meetings were frequently missed or lost, leading to poor follow-through, unclear accountability, and delayed project execution.
- There was no structured way to connect meeting outcomes directly to execution tools — discussions ended at the meeting and rarely translated cleanly into Jira tickets, Azure Board tasks, or Trello cards.
- Past meeting recordings and notes were scattered across tools with no searchable, structured knowledge base — making it difficult to revisit decisions, commitments, or context from previous discussions.
- Data confidentiality requirements meant standard cloud-based transcription and AI tools were unsuitable — the platform needed to support private, on-premise AI processing for sensitive business meetings.
Our Solution
- Built an AI agent system that automatically joins scheduled meetings through calendar and meeting platform integrations, capturing conversations without requiring any manual setup from the user.
- Deployed Whisper for accurate speech-to-text transcription with speaker diarization, producing timestamped, speaker-attributed meeting records that are significantly more useful than raw audio or plain text.
- Integrated LLM processing to automatically generate meeting summaries, formal minutes of meeting, extracted action items, key decisions, blockers, follow-ups, and project-ready task outputs from every transcript.
- Implemented Model Context Protocol (MCP) to connect the AI layer with Jira, Azure Boards, and Trello — enabling the AI agent to create tasks, update boards, and sync action items through standardised tool calls, making integrations reliable, extensible, and easy to maintain.
- Designed a searchable meeting knowledge base using vector embeddings and RAG, enabling teams to query past meeting content with natural language and retrieve relevant decisions, commitments, and context.
Technical Approach
The core engineering challenge on this project was connecting AI outputs to execution tools reliably and maintainably. The naive approach — writing a custom API integration for each project management tool — produces brittle code that breaks whenever third-party APIs change and requires separate maintenance for every new tool added. We chose a different architecture: Model Context Protocol (MCP), which standardises how AI models interact with external tools through a structured layer of defined capabilities and schemas.
By implementing MCP as the integration layer between the LLM and project management tools, we gave the AI agent a standardised way to create tasks, update boards, and sync action items across Jira, Azure Boards, and Trello — through consistent tool calls rather than fragile custom API wrappers. Adding support for a new project management tool becomes a matter of defining its MCP capabilities rather than writing and maintaining a full custom integration. This architectural decision significantly improves the platform's long-term maintainability and extensibility.
The transcription layer uses Whisper deployed for accurate speech-to-text with speaker diarization — producing timestamped, speaker-attributed transcripts that are far more useful for action item extraction and structured documentation than plain text. The LLM processing layer runs over each transcript to generate meeting summaries, formal minutes, extracted action items, key decisions, and blockers. Each output type is structured for both human review in the dashboard and machine-readable consumption by the MCP tool layer.
The searchable meeting knowledge base is built on vector embeddings stored in ChromaDB, with a RAG pipeline that allows teams to query any past meeting using natural language. A question like "what was decided about the Q3 launch timeline?" retrieves the specific meeting segment and decision — without requiring the user to know which meeting it came from or when it occurred. This transforms the accumulation of meeting records from a storage problem into an organisational knowledge asset that compounds in value over time.
Key Features
AI Meeting Agent
Automated AI agents that connect with calendar and meeting platforms to join scheduled meetings, capture conversations, and trigger the full documentation workflow without manual intervention.
Speaker-Attributed Transcription
Accurate meeting transcripts with speaker diarization identifying who said what, with timestamps and clean structure — making post-meeting review fast and reliable.
Structured Meeting Documentation
AI-generated post-meeting outputs including summaries, formal minutes of meeting, key decisions, blockers, follow-up points, and action items — all organised in a single dashboard view.
MCP-Powered Project Tool Integration
Integration with Jira, Azure Boards, and Trello built using Model Context Protocol (MCP) — enabling the AI agent to create tasks, update boards, and sync action items through standardised tool calls rather than brittle custom API wrappers.
Searchable Meeting Knowledge Base
Vector-powered search across all past meeting transcripts and summaries using RAG, allowing teams to query decisions, commitments, and discussion points from any historical meeting.
Technologies Used
Outcomes & Results
- Eliminated manual post-meeting documentation, reducing time spent on meeting admin significantly for teams running multiple meetings per week.
- Improved action item tracking and accountability by connecting meeting commitments directly to Jira, Azure Boards, and Trello as structured tasks.
- Created a fully searchable record of past meeting decisions, discussions, and commitments accessible across the organisation at any time.
- Supported privacy-focused deployments with on-premise AI model processing options for clients with strict data confidentiality requirements.
Want to Build Something Like This?
Let's discuss your project requirements and explore the right approach for your business.
