How ivanpetrushev/meeting-notes-extractor works
Main Components
a) Input Layer:
• Input S3 Bucket (for raw recordings)
• Input Plugins System
• Google Drive Plugin (implemented)
• Google Meet Bot Plugin (planned)
• Future plugins placeholder
b) Processing Layer:
• AWS Lambda Functions
• start_transcribe Lambda
• summarize Lambda
• AWS Services
• AWS Transcribe
• AWS Bedrock (Claude 3 Haiku)
• Intermediate S3 Bucket (for transcripts)
c) Output Layer:
• Output S3 Bucket (for processed notes)
• Output Plugins System
• Google Chat Plugin (implemented)
• Future plugins placeholder
[Input Sources] → [Input Plugins] → [Core Processing] → [Output Plugins] → [Output Destinations]
Key Elements to Highlight
• Rectangles: AWS Lambda functions
• Hexagons: AWS services (Transcribe, Bedrock)
• Rounded rectangles: Plugins
• Clouds: External services (Google Drive, Google Chat)
• Blue: AWS native services
• Green: Input components
• Orange: Processing components
• Purple: Output components
• Gray: Future/planned components
Following flows:
• Recording upload → Input S3
• Input S3 → start_transcribe Lambda
• start_transcribe Lambda → AWS Transcribe
• AWS Transcribe → Transcript S3
• Transcript S3 → summarize Lambda
• summarize Lambda → AWS Bedrock
• summarize Lambda → Output S3
• Output S3 → Output plugins
Important Details
• Infrastructure elements:
• Terraform managed components
• DynamoDB for state management
• IAM roles and permissions flow
• Plugin Architecture:
• Plugin extension points
• Demonstrate plugin independence
• Configuration flow
Technical Notes
• Technologies:
• Node.js (visible in package.json files)
• Terraform for IaC
• Docker for bot components
• AWS SDK integrations
Recommendations
• Brief annotations for complex flows
• Error handling paths
• Monitoring points