Promo Miner: Shopping AI Agent

An Autonomous Gmail-Based Deal Discovery and Comparison Agent
AI Agents
Award Winner
Deployment
Large Language Models
Monitoring
Natural Language Processing
UI/UX
Web Development
Author

Nivedhitha Dhanasekaran

Published

April 13, 2025

Project Overview

Promo Miner is an autonomous AI-powered shopping assistant that uncovers hidden savings in users’ Gmail inboxes. It identifies promotional offers from the Gmail Promotions tab, uses LLMs to rank deal quality, and performs real-time web comparisons to ensure users always get the best available deal.

✨ Built a fully integrated shopping agent that autonomously performs Gmail parsing, promo reasoning, cross-site comparison, and Chrome-based alerting in real time.
✨ Combines Gemini LLM reasoning with Exa-based search and push notifications to deliver ranked savings with minimal user effort.

Description

The Promo Miner system securely accesses only the Gmail Promotions tab using OAuth and filters out irrelevant emails like welcome messages. Valid promotional content is parsed and structured, extracting metadata such as discounts, brand names, expiration dates, and coupon codes. These are stored in Firestore for persistent deal tracking.

A Gemini-based LLM agent then reasons over this data to classify and rank deals based on urgency, value, and usefulness. Deals that are likely to expire soon or offer significant savings are highlighted. The system also employs an Exa-powered agent to compare prices across platforms like Amazon. If a better offer is found, the user receives a real-time push notification through a Chrome Extension interface.

Download PDF file.

1. Email Access & Filtering
  • Uses Gmail API with OAuth 2.0 to securely access the Promotions tab
  • Filters out onboarding/welcome emails and identifies valid promotional content
  • Parses subject line, sender, body text, promo code, and expiration cues
2. Data Storage & Processing
  • Extracted metadata is stored in Google Firestore for persistent tracking
  • Supports longitudinal deal analysis and browser-based rendering
  • Integrates with Chrome Extension for lightweight front-end access
3. LLM-Based Promo Reasoning
  • Gemini LLM classifies each deal on urgency, product type, savings, and utility
  • Generates user-specific promo rankings using context-aware heuristics
  • Annotates whether a deal includes coupons, bundle offers, or price drops
4. Web Comparison Agent
  • Exa search agent compares extracted deals to live listings on platforms like Amazon
  • Flags cheaper alternatives or higher-value bundles
  • Triggers push notifications to users for better savings
5. Deployment & Interface
  • Chrome Extension provides a ranked deal dashboard
  • Users receive timely alerts on new or expiring promos
  • Entire pipeline runs in real time, with low-latency triggers from backend to browser

Tools & Frameworks

Component Stack / Tools Used
Email Access Gmail API, OAuth 2.0
Parsing & Storage Flask, Google Firestore, Regex, JSON Extractors
Reasoning Agent Gemini, Python, Structured Query Prompting, LlamaIndex
Web Comparison Agent Exa, Semantic Search, Jina Reader API
Frontend Interface Chrome Extension, JavaScript, HTML, CSS, Browser Notifications API
Back-end Infrastructure Flask REST API, Firestore Rules