Ripple: Voice-to-Action Meeting Assistant

Inspiration

Meetings are full of decisions, action items, and deadlines, but important commitments are often forgotten or lost in chat messages. We built Ripple to bridge the gap between what teams discuss and what they actually get done.

What it does

Ripple transforms short meeting recordings into candidate tasks with speaker labels, transcript timestamps, suggested owners, deadlines, and supporting transcript evidence. Team members can review the suggestions, map anonymous speakers to people, edit task details, and approve or reject each suggestion.

Only approved tasks are saved. Once saved, teams can assign tasks, track their status, and revisit their commitments through a task dashboard.

How we built it

We built the frontend using React, TypeScript, and Vite. A Python FastAPI service coordinates audio processing: AssemblyAI provides transcription and speaker diarization, while an OpenRouter-hosted language model extracts candidate tasks.

The frontend presents these suggestions for human review. Approved tasks are sent through an MCP (Model Context Protocol) client to our Node.js and TypeScript MCP server. The server exposes structured tools for creating, listing, assigning, and updating task status. SQLite provides persistent storage.

Challenges we tackled

One key challenge was keeping AI-generated suggestions separate from saved tasks. We designed the workflow so that a person reviews the suggestions before the frontend writes approved tasks through MCP. We also worked on connecting the Python voice service, React frontend, MCP server, and SQLite database into one testable workflow.

What we learned

Building Ripple helped us understand how MCP can expose application capabilities as reusable tools with structured inputs and results. We also learned that useful AI automation needs validation, transparency, and human oversight—not just model-generated output.

What's next

We plan to improve deadline extraction and evaluation, strengthen the production-readiness of the application, and explore integrations with team collaboration tools. Ripple is currently a working local prototype, and live meeting analysis requires configured provider API keys.

Built With

• ai
• assemblyai
• automation
• express.js
• fastapi
• llm
• mcp
• node.js
• openrouter
• python
• react
• sppech-to-text
• sqlite
• typescript
• vite