DEV Community

Cover image for Building Meridian: An Autonomous Multi-Agent AI Scheduler with Gemini 3.1
Mohammed Ayaan Adil Ahmed
Mohammed Ayaan Adil Ahmed

Posted on

Building Meridian: An Autonomous Multi-Agent AI Scheduler with Gemini 3.1

πŸš€ The Vision: Beyond Static Scheduling

Scheduling meetings is a universal friction point. We've all experienced the "email tag" fatigue and the loss of context once a meeting ends. For the Google Cloud Gen AI Academy APAC Edition, my teammate Bibi Sufiya Shariff and I built Meridian. An autonomous multi-agent system designed to handle the entire meeting lifecycle.

🧠 The Architecture: Orchestrator & Sub-Agents

Meridian isn't a single script; it’s a fleet of specialized agents coordinated by a central "brain."

  • The Orchestrator (Gemini 3.1): Using Vertex AI, the orchestrator parses natural language intent and dynamically delegates tasks to sub-agents based on the context.
  • Calendar Agent: Interfaces with the Google Calendar API to surface real-time availability.
  • Email Agent: Handles automated dispatch of invites and notifications.
  • Transcription & Summary Agents: Processes audio/text post-meeting to extract action items and summaries.

πŸ’Ž The USP: "Glass Box" Transparency

A major challenge with AI agents is trust. Users often feel like they are interacting with a "black box."

We solved this by implementing a Real-time Agent Trace. Using Server-Sent Events (SSE), Meridian streams the agent’s internal reasoning, tool calls, and state changes directly to the UI. You can watch the AI "think" and "act" in real-time.

πŸ› οΈ The Tech Stack

  • Frontend: Next.js (App Router) for a premium, high-aesthetic dashboard.
  • Backend: FastAPI (Python) serving as the Agent Hub.
  • AI Layer: Google Vertex AI (Gemini 3.1).
  • Persistence: Google Cloud SQL (PostgreSQL).
  • Hosting: Google Cloud Run for scalable, containerized deployment.

πŸ“ˆ Roadmap & Future Scope

This is just the beginning. We're looking forward to expanding Meridian with:

  1. Collaborative Workspaces: Multi-user calendar comparison for team-wide scheduling.
  2. Deep Memory: Using persistent agent state to remember context across months of meetings.
  3. Bio-rhythm Optimization: Suggesting slots based on user productivity patterns.

Building Meridian has been an incredible journey in exploring the limits of agentic workflows. A huge thank you to the Google Cloud team for the support!


πŸ”— Stay Connected

Check out the project and let us know what you think in the comments! πŸ‘‡

machinelearning #productivity #webdev #googlecloud

Top comments (0)