AI Meeting Notes Agent

AI Meeting Notes Agent automatically summarizes meetings, extracts key action items, and sends clean reports with PDF summaries — multilingual and powered by AWS Bedrock.

Comment

🧠 About the Project: AI Meeting Notes Agent

🎯 Inspiration

Every day, countless hours of meetings are spent discussing key ideas, decisions, and next steps — yet most of this knowledge is lost in messy transcripts or forgotten notes.

We wanted to build an autonomous AI agent that can listen, summarize, and deliver actionable insights from meetings automatically.

Goal: save time, reduce cognitive load, and help teams focus on what matters most — taking action.

⚙️ How We Built It

The AI Meeting Notes Agent was built as a serverless application using AWS services and Anthropic’s Claude 3 Sonnet model through Amazon Bedrock.

🧩 Architecture Overview

  1. Meeting transcript (or audio transcription) is stored in Amazon S3.
  2. A Lambda function triggers automatically when a new transcript is uploaded.
  3. The function calls Amazon Bedrock (Claude 3 Sonnet) using the Messages API for reasoning.
  4. The agent:
  • Summarizes the meeting
  • Detects the language (English or French)
  • Extracts structured Action Items
  1. The output is formatted, converted into a PDF with ReportLab, and sent via Amazon SES to participants.
  2. A local test version (local_test.py) enables quick iteration before deployment.

Architecture Diagram

🧰 Technical Stack

Component Purpose
AWS Bedrock Reasoning LLM (Claude 3 Sonnet)
AWS Lambda Serverless agent logic
Amazon S3 Transcript storage
Amazon SES Email delivery (PDF attachment)
Python 3.13 Backend logic
ReportLab PDF generation
Draw.io Architecture diagram

🧠 What We Learned

  • Structuring reasoning prompts for Claude 3 Sonnet using the new Messages API
  • Managing serverless orchestration across Bedrock, S3, and SES
  • Optimizing LLM outputs for multilingual clarity
  • Generating professional PDF reports from unstructured data
  • Simplifying user experience — one upload = one meeting summary

⚔️ Challenges

  • Understanding Bedrock’s new Messages API and adapting old prompt formats
  • Handling IAM permissions for Bedrock and SES
  • Ensuring formatting consistency (plain text → HTML → PDF)
  • Maintaining low latency for quasi real-time processing

$$ \text{Latency} \approx \text{Input Size} \times 0.2s + \text{Model Response Time} $$

🚀 What’s Next

  • 🔊 Add AWS Transcribe for automatic audio-to-text
  • 💬 Integrate with Slack / Microsoft Teams for instant summaries
  • 📋 Connect to Notion, Trello, or Zapier for auto task creation
  • 💼 Package it as a SaaS template for teams and consultants

💡 Key Takeaway

AI Meeting Notes Agent transforms raw meeting data into structured, actionable knowledge. It’s a real-world example of how Amazon Bedrock’s agentic AI can bridge the gap between conversation and execution.

Built With

Share this project:

Updates