AI
Meeting Intelligence Assistant
Multi-Agent Orchestration • Microsoft 365 • Teams Toolkit • TypeScript

A modern AI assistant that joins meetings, understands conversations, and takes action automatically.

This project is a real-time, multi-agent meeting intelligence system built in TypeScript for Microsoft Teams. It can check inbox emails, manage calendar activity, join meetings, answer live questions about the conversation, generate meeting minutes, and automatically send follow-up summaries to attendees after the meeting ends.

Get Started Watch Demo
5+ Specialized agents
Live Meeting reasoning
M365 Tool integration
Auto Minutes and follow-ups
Demo

Project walkthrough video

Embed your YouTube demo here after uploading your recording.

Overview

What this project is

Traditional meeting tools stop at transcription. This project goes further by combining multiple AI agents, Microsoft 365 integrations, and orchestration logic to turn meetings into actionable intelligence.

The problem

Meetings are where decisions are made, but they are also where context gets lost. Manual notes are inconsistent, action items are forgotten, and following up takes extra time. People often leave a meeting without a clear record of what was decided, who owns what, and what was said.

The solution

This assistant acts as a real-time meeting companion and post-meeting automation engine. It listens, understands, retrieves context, summarizes outcomes, and sends follow-up communication automatically so that meetings lead to clear outcomes instead of scattered notes.

Capabilities

Key features

The system combines meeting intelligence, productivity workflows, and automation into one coordinated experience.

Real-time meeting intelligence

Understands live conversation flow and supports contextual questions during the meeting itself.

Live context Speaker-aware Fast response

Inbox and calendar productivity

Checks emails, reviews calendar events, and helps prioritize productivity-related tasks using Microsoft 365 tools.

Email access Calendar access Task support

Automated meeting minutes

Creates structured notes including decisions, discussion points, owners, and next steps after the meeting ends.

Summaries Action items Minutes

Automatic follow-up emails

Sends post-meeting communication to attendees with summaries and action items, reducing manual effort.

Attendee sync Email automation Clear outputs

Multi-agent orchestration

Uses specialized agents working together instead of relying on one generic assistant to do everything.

Orchestrator Delegation Scalable design

Teams-native workflow

Built around the Microsoft Teams ecosystem using the M365 toolkit and TypeScript for practical enterprise use.

Teams TypeScript Microsoft 365
Architecture

How the system is structured

This solution is designed as a coordinated multi-agent system, where each agent focuses on a specific responsibility.

1

Orchestrator Agent

Routes user requests, coordinates the workflow, selects the right specialized agent, and manages overall task execution.

2

Meeting Intelligence Agent

Handles live meeting context, real-time understanding, speaker-related questions, and meeting state awareness.

3

Summarization Agent

Converts raw discussion into clear minutes, decisions, highlights, and action items after the call.

4

Email Agent

Creates and sends follow-up emails to attendees with structured post-meeting outputs.

5

Productivity Agent

Handles inbox checks, calendar lookups, prioritization tasks, and Microsoft 365 productivity actions.

This design makes the solution more reliable and scalable because each agent is responsible for a focused function instead of sharing one overloaded prompt.
Repository

Suggested project structure

Adjust the folders below to match your actual codebase, but this format works well for a Teams Toolkit TypeScript project.

project-root/
├── src/
│   ├── agents/
│   │   ├── orchestrator/
│   │   ├── meeting-agent/
│   │   ├── summary-agent/
│   │   ├── email-agent/
│   │   └── productivity-agent/
│   ├── services/
│   │   ├── graph/
│   │   ├── meetings/
│   │   ├── mail/
│   │   └── calendar/
│   ├── utils/
│   ├── config/
│   └── index.ts
├── appPackage/
├── env/
├── package.json
├── tsconfig.json
├── README.md
└── docs/
Before you start

Prerequisites

Make sure the environment below is ready before running the application.

Development requirements

  • Node.js 18 or later
  • npm or pnpm
  • Visual Studio Code
  • Microsoft Teams Toolkit extension
  • Microsoft 365 developer tenant or organization tenant access

Cloud and integration requirements

  • Microsoft 365 app registration and permissions
  • Teams app configuration
  • Email and calendar access through Microsoft Graph
  • Any AI model or orchestration service your agents depend on
  • Meeting join permissions where applicable
Setup

How to configure the project

Use the steps below as your technical walkthrough section for the buildathon submission.

1

Clone the repository

Pull the project locally so you can install dependencies and configure the environment.

git clone YOUR_REPO_URL
cd YOUR_PROJECT_FOLDER
2

Install dependencies

Install all required packages for the Teams app, agent orchestration, and Microsoft 365 integrations.

npm install
3

Configure environment variables

Create your environment file and add all required keys, IDs, secrets, and configuration values.

# Example .env values
M365_CLIENT_ID=
M365_CLIENT_SECRET=
M365_TENANT_ID=
TEAMS_APP_ID=
GRAPH_SCOPES=
AI_MODEL_ENDPOINT=
AI_MODEL_KEY=
EMAIL_SENDER=
MEETING_SERVICE_CONFIG=
4

Set up Microsoft 365 and Teams permissions

Make sure the app registration has the permissions required for mail access, calendar access, meeting participation, and any other Microsoft Graph operations your solution uses.

5

Update Teams Toolkit configuration

Verify the Teams app manifest, local environment settings, and any toolkit deployment profiles.

6

Start the project

Run the local dev server and launch the Teams app for testing.

npm run dev
Run flow

How the project works at runtime

This section explains the end-to-end experience from the moment a meeting starts to the final follow-up email.

Live meeting flow

  1. User opens the Teams app or initiates the assistant.
  2. The orchestrator agent identifies the request context.
  3. The system joins or monitors the active meeting.
  4. Meeting intelligence agent listens and tracks live discussion context.
  5. The user can ask questions such as who said what during the meeting.

Post-meeting flow

  1. The meeting ends and the summary workflow is triggered.
  2. The summarization agent generates minutes and action items.
  3. The email agent prepares a follow-up message.
  4. Attendees receive the minutes automatically.
  5. The record can be stored or extended into future workflows.
If your assistant can answer “What did John say?” during the meeting, mention in your video that this is real-time contextual retrieval, not just post-meeting summarization.
Commands

Useful development commands

Replace these with the exact scripts from your package.json if they differ.

Command Purpose
npm install Installs all project dependencies
npm run dev Starts the app in development mode
npm run build Builds the TypeScript project for production
npm run lint Runs lint checks for code quality
npm run test Runs the test suite if configured
Use cases

Where this assistant delivers value

This project is practical for internal enterprise workflows and not just a concept demo.

Executive meetings

Captures decisions, owners, and next steps without relying on manual note-taking.

Project standups

Keeps a record of blockers, decisions, and action items for faster accountability.

Client calls

Preserves meeting context and automates follow-up summaries after stakeholder discussions.

Troubleshooting

Common issues and fixes

Use this section in your submission to show that the project is practical and implementation-aware.

App does not start

  • Check Node.js version compatibility
  • Run a clean install of dependencies
  • Verify TypeScript build errors
  • Confirm environment variables are present

Teams integration issues

  • Verify Teams Toolkit configuration
  • Check manifest values and app IDs
  • Ensure app permissions are granted
  • Confirm tenant sign-in and consent are complete

Email or calendar calls fail

  • Check Microsoft Graph scopes
  • Verify client secret and tenant settings
  • Confirm the correct signed-in identity is being used

Meeting data is incomplete

  • Confirm meeting join flow is working
  • Check transcription or live context source
  • Make sure the meeting agent is receiving events correctly
Submission notes

What to highlight in your buildathon entry

These are the strongest points to emphasize when judges review your repository and video.

  • This is a real multi-agent system, not a single prompt pretending to be one
  • The assistant uses Microsoft 365 integrations for practical enterprise workflows
  • It supports live contextual meeting questions while the meeting is still in progress
  • It produces structured minutes and automatically follows up afterward
  • It demonstrates clear end-to-end value from conversation to action