InboxPilot – Autonomous AI Email Triage Assistant

InboxPilot is an autonomous AI email triage assistant that classifies emails, prioritizes urgency, drafts responses, syncs actionable items to Notion, and learns user preferences from feedback.

Comment

Inspiration

Email overload is a common problem for both students and professionals. Important messages often get buried among newsletters, notifications, and low-priority communication. Manually reviewing every email consumes time and attention, while delayed responses can result in missed opportunities.

I wanted to build an autonomous AI agent that could reduce this cognitive burden by understanding incoming emails, identifying what matters, generating actionable insights, and helping users respond faster.

What it does

InboxPilot is an autonomous AI email triage assistant powered by Google Gemini 3.6 Flash.

The system continuously analyzes incoming emails and performs intelligent triage by:

  • Retrieving emails from Gmail
  • Classifying category and priority
  • Detecting urgency and spam signals
  • Generating concise summaries
  • Creating AI-powered reply drafts
  • Synchronizing actionable information to Notion
  • Learning user preferences through feedback

Instead of acting as a simple classifier, InboxPilot operates as an agent that reasons about incoming information and decides how it should be processed.

How we built it

InboxPilot was built using:

  • Python 3.11
  • FastAPI
  • Google Gemini 3.6 Flash
  • Google GenAI SDK
  • Gmail API
  • Notion API
  • Docker
  • Google Cloud Run
  • Pydantic

The backend follows a modular architecture consisting of:

  • Gmail Service for email retrieval and draft creation
  • AI Agent Layer for reasoning and classification
  • Email Triage Workflow for orchestration
  • Memory Service for preference learning
  • Notion Integration Layer for workspace synchronization
  • FastAPI API Layer for serving endpoints

The application was containerized using Docker and deployed to Google Cloud Run.

Challenges we ran into

Several challenges emerged during development:

  • Designing reliable AI classification prompts
  • Handling Gmail OAuth authentication securely
  • Preventing duplicate processing of emails
  • Building preference-aware behavior using user feedback
  • Managing integrations across Gmail, Gemini, Notion, and Cloud Run
  • Debugging deployment and environment configuration issues

A major challenge was ensuring the agent produced consistent classifications while remaining flexible enough to adapt to user corrections.

Accomplishments that we're proud of

I am proud that InboxPilot successfully combines multiple systems into a cohesive autonomous workflow.

Key accomplishments include:

  • End-to-end Gmail email processing
  • AI-powered triage and prioritization
  • Automatic draft generation
  • Notion synchronization
  • Preference-aware learning through feedback
  • Dockerized deployment
  • Successful deployment on Google Cloud Run

The project demonstrates how modern AI agents can move beyond simple chat interactions and actively assist with real productivity workflows.

What we learned

During development we gained hands-on experience with:

  • Agent-oriented system design
  • Prompt engineering for classification tasks
  • Google Gemini integration
  • Gmail API workflows
  • Notion API integration
  • Cloud-native deployment using Docker and Cloud Run
  • Building maintainable service-oriented architectures in FastAPI

I also learned that deploying and integrating production-style systems is often more challenging than implementing individual features.

What's next for InboxPilot – Autonomous AI Email Triage Assistant

Future improvements include:

  • Multi-user authentication
  • Long-term preference memory
  • Calendar integration
  • Task extraction and scheduling
  • Advanced workflow automation
  • Team collaboration support
  • Improved reasoning using richer contextual memory

Our long-term vision is to evolve InboxPilot into a complete AI-powered productivity agent that can manage communication and action items across multiple platforms.

Built With

  • agentic-ai
  • ai
  • dockers
  • fastapi
  • gemail-api
  • google-cloud-run
  • google-gemin
  • google-genai-sdk
  • machine-learning
  • notion-api
  • pydantic
  • python
Share this project:

Updates

Submission history