Efesto

Inspiration

Efesto: Multi-Agent Meeting Intelligence Platform

Inspiration

Modern teams spend countless hours in meetings, but much of the knowledge, action items, and insights are lost or underutilized. We wanted to build a platform that not only manages meetings but also leverages AI agents to research, analyze, and summarize meetings—turning every session into actionable intelligence.

What It Does

Efesto is a cloud-native, AI-powered meeting management API that enables teams to:

• Create, schedule, and manage meetings with rich metadata (participants, agenda, tags, etc.)
• Upload and process audio recordings for transcription and analysis
• Trigger AI research agents when meetings start, providing real-time context, research, and recommendations
• Automatically generate comprehensive final reports when meetings end, including executive summaries, action items, and insights
• Integrate with any frontend (web/mobile) via a robust, documented REST API

How We Built It

• Backend: Python (FastAPI), SQLAlchemy, Pydantic
  
• Database: PostgreSQL (via Supabase)
  
• Cloud Platform: Google Cloud Run (Dockerized), Cloud Build, IAM, Logging
  
• AI/ML: OpenAI GPT-4 for research agents and report generation, Whisper for transcription
  
• DevOps: Docker, CI/CD scripts, environment variable management
  
• Documentation: Markdown docs, OpenAPI/Swagger for live API docs
  

Architecture Diagram

Diagram not included in markdown.

Data Sources

• Primary: PostgreSQL database (meetings, users, audio metadata)
  
• Audio: Cloud Storage (audio files)
  
• AI: OpenAI APIs (GPT-4 for research/summary, Whisper for transcription)
  
• External: (Optional) Calendar/email integrations for future expansion
  

Features & Functionality

• Full CRUD for meetings (create, read, update, delete)
• Audio file upload/download and transcription
• AI research agents that activate when meetings start
• Automatic final report generation when meetings end
• Comprehensive API documentation and health endpoints
• Production-ready deployment with monitoring, scaling, and security

Challenges We Ran Into

• Cloud Build/Run region limitations (solved by switching to us-east1)
• ENUM validation and field mapping for dynamic meeting schemas
• Dependency conflicts (Google ADK, SQLAlchemy, etc.)
• Ensuring robust error handling for all endpoints
• Coordinating AI agent triggers with meeting lifecycle events

Accomplishments That We're Proud Of

• Fully automated, production-grade deployment on Google Cloud Run
• Dynamic field mapping for flexible meeting schemas
• Seamless AI integration for research and reporting
• Comprehensive, developer-friendly documentation
• Ready-to-use API for any frontend team

What We Learned

• Cloud-native best practices for scalable, secure APIs
• How to orchestrate AI agents in real-world workflows
• Importance of robust documentation for cross-team handoff
• Effective troubleshooting of cloud build and deployment issues
• How to deliver business value with actionable meeting intelligence

What's Next for Efesto

• Real-time collaboration features (live notes, chat, Q&A)
• Deeper integrations (calendar, email, task managers)
• Advanced analytics (meeting effectiveness, participation trends)
• Custom AI agents for different meeting types
• Enterprise security and compliance enhancements

Technologies Used

• Python, FastAPI, SQLAlchemy, Pydantic
• PostgreSQL, Supabase
• Google Cloud Run, Docker, Cloud Build
• OpenAI GPT-4, Whisper
• Markdown, OpenAPI/Swagger

Findings & Learnings

• AI can dramatically enhance meeting productivity by surfacing insights and automating summaries.
• Cloud-native deployment ensures scalability and reliability.
• Dynamic schema mapping is essential for flexible, future-proof APIs.
• Clear documentation and modular architecture are key for rapid team onboarding and handoff.

Built With

• cloud-build-openai-gpt-4
• docker
• fastapi
• pydantic-postgresql
• python
• sqlalchemy
• supabase-google-cloud-run
• whisper-markdown