How I Built an AI-Powered Workday Assistant with Gemini, Gmail & Google Calendar in One Day

#googleaichallenge #nextjs #gemini #promptwars

The Problem I Set Out to Solve

Every professional I know drowns in the same two things: emails and calendar conflicts.
We spend the first hour of every workday just figuring out what actually matters.
I wanted to fix that with AI.

What I Built: FlowDesk

FlowDesk is an AI-powered workday assistant that connects your Gmail and Google Calendar,
then uses Gemini AI to tell you exactly what matters today — in plain English.

Live demo: https://flowdesk-914324693806.us-central1.run.app

GitHub: https://github.com/TanushReddy-Dev/Flowdesk

The Tech Stack

Next.js 14 (App Router + TypeScript)
Gemini API — AI reasoning, summarization, chat
Gmail API — fetch and summarize inbox
Google Calendar API — today's events + conflict detection
Google OAuth 2.0 — one-click sign in
Google Cloud Run — serverless deployment
NextAuth.js — session management

How It Works

User signs in with Google OAuth — zero setup
FlowDesk fetches the last 15 unread emails + today's calendar events server-side
Everything is sent to Gemini with an executive assistant system prompt
Gemini returns a structured Daily Briefing: priorities, conflicts, suggested focus
User can chat naturally: "Do I have time for lunch?" or "What's my most urgent email?"
User can generate AI email drafts with one click

The AI Decision Logic

The core insight is context stacking. Instead of asking Gemini one thing at a time,
FlowDesk sends the full picture — all emails, all events, current time — in a single
structured prompt. Gemini reasons across everything simultaneously and returns a
prioritized, conflict-aware briefing.

User data (emails + calendar + time)
↓
Gemini AI reasoning
↓
{ summary, priorities, conflicts, suggestedFocus }
↓
Daily Briefing card

The Biggest Challenge

Getting Gmail and Calendar data to arrive together before rendering the briefing.
I solved this with parallel Promise.all() calls on the server side, keeping the
dashboard load under 2 seconds even with multiple API calls.

What I Learned Building with Google Antigravity

Antigravity completely changed how I think about building. Instead of writing code
line by line, I described what I wanted and the agent planned, scaffolded, and built
entire modules autonomously. The hardest part was writing good prompts — being specific
about file structure, error handling, and TypeScript strictness made the difference
between generic output and production-quality code.