EchoSync

Inspiration

Meetings are the lifeblood of collaboration, but they are often black holes for information. We've all asked, "What did we decide three days ago?" EchoSync was built to ensure that every insight, decision, and action item is captured, reasoned about, and remembered.

What it does

EchoSync is a real-time voice agent that does the heavy lifting for you:

• Diarized Transcription: Distinguishes between speakers instantly using Amazon Nova 2 Sonic.
• Agentic Reasoning: Uses Nova 2 Lite to extract action items, track sentiment, and answer live questions about meeting history (RAG).
• Multimodal Analysis: Analyzes whiteboards or charts mid-meeting with Nova Vision.
• Active Automation: One-click email summaries via Amazon SES and calendar integration via Nova Act.

How I built it

The backend is powered by FastAPI and Boto3, leveraging the full suite of Amazon Nova models in AWS Bedrock. We used DynamoDB for persistent meeting memory and S3 for storage. The frontend is a modern React dashboard providing a low-latency AI experience through WebSockets.

Challenges I ran into

Handling real-time speaker diarization without massive compute was a challenge. We solved this by combining the ultra-low latency of Nova 2 Sonic with a custom gap-detection heuristic to keep the experience fast and cost-effective.

Accomplishments that I am proud of

The Live Q&A feature is a game-changer. Being able to ask the agent about a meeting that happened weeks ago while you are still in a live call feels like magic and significantly improves team alignment.

What's next for EchoSync

We plan to expand with deeper MCP (Model Context Protocol) integrations, allowing the agent to proactively update Jira tickets and Slack channels directly from the conversation.

Built With

• amazon-nova-2-lite
• amazon-nova-2-sonic
• aws-bedrock
• aws-dynamodb
• aws-ses
• fastapi
• nova-act
• nova-vision
• python
• react
• websockets