-
-
AI Productivity System Architecture Overview
Inspiration
Managing college life is chaos — lectures, assignments, DSA practice, hackathons all competing for attention. I wanted an AI that actually knows my schedule and helps me plan intelligently, not just a generic to-do app.
What it does
FlowMind is an AI productivity manager you talk to naturally. It reads your Google Calendar, manages your tasks, and helps you plan your day. Ask it "What's on my schedule today?" or "Block 2 hours for my assignment tonight" and it actually does it — using real calendar data and live task management.
How I built it
- Gemini 2.0 Flash as the AI brain with real-time function calling
- Google GenerativeAI Python SDK for the agentic loop
- FastAPI backend with 6 custom tools: get today's events, upcoming events, get/create/complete tasks, suggest schedule blocks
- React + Vite + Tailwind CSS for the frontend
- Web Speech API for voice input
- Railway for deployment
Challenges I ran into
Getting the Gemini function calling loop to work reliably with multiple tool calls in a single conversation turn. Also handling Python version compatibility (Python 3.14 broke pydantic — had to use 3.11).
What I learned
How to build a real agentic AI system with Gemini's function calling — where the AI autonomously decides which tools to call, executes them, and synthesizes the results into a helpful response.
What's next
Real Google Calendar OAuth so users can connect their actual calendar, and a mobile app version.
Log in or sign up for Devpost to join the conversation.