Logo for the website. A blue P-like symbol and a red K-like symbol. Pushkar Kurhekar 
Back to projects
•
3 min read

MailMate - AI Email Assistant

Table of Contents

About

MailMate is an AI-powered email assistant that helps professionals manage email overload through intelligent automation. The Chrome extension integrates directly with Gmail to provide instant email summarization, automatic action item extraction, and contextual reply generation powered by Google’s Gemini models.

Features

  • Smart Email Summarization: Quickly understand key points from long email threads
  • Action Item Extraction: Automatically identify tasks, deadlines, and follow-ups
  • Contextual Reply Generation: Create professional responses with one click
  • Gmail Integration: Seamless Chrome extension that works within your inbox
  • Quality Assurance: Advanced ranking system ensures high-quality AI outputs

Technology Stack

  • AI/ML: Vertex AI (Gemini 1.5 Flash), Custom prompt engineering
  • Backend: Flask, PostgreSQL, Cloud Run, Cloud SQL
  • Data Pipeline: Apache Airflow, Great Expectations, DVC
  • Frontend: Chrome Extension (JavaScript/HTML/CSS)
  • Infrastructure: Google Cloud Platform, Docker, GitHub Actions

Implementation

MailMate uses a multi-pipeline architecture designed for scalability and reliability:

Data Pipeline

  • Orchestration: Apache Airflow manages the complete data workflow, from email ingestion to preprocessing
  • Quality Control: Great Expectations validates data integrity at each pipeline stage
  • Version Control: DVC tracks dataset versions and ensures reproducibility
  • Processing: Automated cleaning and standardization of email formats for optimal model performance

Model Pipeline

  • LLM Integration: Leverages Vertex AI’s Gemini 1.5 Flash model with custom prompt engineering
  • Ranking System: Implements a multi-generation approach where the model produces multiple responses, which are then evaluated and ranked based on quality metrics
  • Prompt Optimization: Iterative prompt refinement based on performance metrics and user feedback
  • MLflow Tracking: Comprehensive experiment tracking for prompt variations and model performance

Deployment Architecture

  • Backend Services: Flask API deployed on Google Cloud Run for auto-scaling and high availability
  • Database: Cloud SQL (PostgreSQL) for secure storage of user preferences and interaction logs
  • CI/CD: GitHub Actions automates testing, building, and deployment workflows
  • Monitoring: Integrated logging and performance tracking across all services

Key Technical Achievements

  • Response Quality: The ranking system improved response relevance by comparing multiple generated outputs and selecting the best one based on coherence, completeness, and context alignment
  • Scalability: Serverless architecture on Cloud Run handles varying loads efficiently with automatic scaling
  • Security: Implemented secure OAuth2 authentication for Gmail access with minimal permission scope
  • Performance: Optimized prompt engineering reduced average response generation time while maintaining quality

User Workflow

  1. Installation: Users install the Chrome extension from the Chrome Web Store
  2. Authentication: Secure OAuth2 flow grants minimal required permissions for Gmail access
  3. Usage: Within Gmail, users can click the MailMate icon to:
    • Get instant summaries of long email threads
    • Extract all action items and deadlines with one click
    • Generate contextually appropriate reply drafts
  4. Customization: Users can adjust tone preferences and response length for generated replies