automail

Inspiration

I built Automail because I was overwhelmed. My inbox had turned into a chaotic mess of newsletters, follow-ups, receipts, and things I kept forgetting to reply to. I didn’t want to manually label and dig through threads every morning. I wanted an AI-powered assistant that could read my emails, understand them, and help me stay on top of everything without the daily grind.

What it does

Automail connects to your Gmail and acts as an intelligent assistant for your inbox. It:

• Automatically labels emails using AI-powered semantic tagging
• Extracts todos, reminders, and deadlines from email content
• Generates smart draft replies using Claude AI, based on tone and intent
• Saves reminders and tasks to a local or cloud database for follow-up
• Displays everything in a clean dashboard with customizable widgets like “Follow Up,” “Finance,” or “Drafts” ## How we built it I built Automail using a modern, full-stack setup. On the backend, I used FastAPI with Python and Uvicorn to handle API routing and Gmail integration. Emails are fetched using the Gmail API, processed with Claude AI via the Anthropic SDK, and stored in PostgreSQL for structured entities like reminders and MongoDB for more flexible logs and metadata. On the frontend, I used React 18, Vite, TypeScript, and Tailwind CSS to create a clean, responsive interface. The app is modular, with customizable bento-style widgets that update dynamically based on processed email data.

Challenges we ran into

Gmail’s API, especially around threading and MIME parsing, was more complex than expected. Draft replies needed to be inserted with proper threading headers to avoid creating new threads in Gmail. Another challenge was designing prompts that made the AI both accurate and safe, especially when extracting reminders or deciding whether a reply was needed. Finally, syncing the database state with what’s happening in Gmail without duplication or mismatch required careful logic and idempotent database operations. A big challenge was having push notifications from Gmail by setting up a Pub/Sub service on GCP which only works with https hosted websites/services.

Accomplishments that we're proud of

I'm proud that Automail not only connects to Gmail but actually understands emails in a meaningful way. It can tag them intelligently, extract real tasks and deadlines, and even draft context-aware replies. Getting everything to work smoothly, Gmail, AI, frontend widgets, and databases; felt like orchestrating multiple moving parts. Seeing the system process a new email end-to-end in real time and surface the result in a clean dashboard was a huge milestone. Also open source and complete control over my email!

What we learned

I learned how powerful AI can be when combined with deterministic systems. LLMs like Claude are excellent at extracting meaning, but structure and consistency still come from databases and clear logic. I also learned that UI and UX matter deeply, an AI assistant is only helpful if the user trusts it and feels in control. Finally, we gained a deeper appreciation for how email works under the hood, especially when building reliable, production-grade integrations. I also got better at vibe-coding frontend!

What's next for automail

More powerful, more AI insights and local-LLM support for privacy first email client.

Built With

• anthropic
• claude
• fastapi
• python
• v0
• vercel