CouponAI: Turn Promotional Emails into Smart Savings

This is a submission for the Postmark Challenge: Inbox Innovators.

What I Built

I built CouponAI, a server-side application that automates the extraction of promotional offers, deals, and coupons from your inbox using Postmark’s inbound email parsing feature.

Many of us receive dozens of marketing and promotional emails every day from services like Swiggy, Zomato, Uber, Ola, e-commerce brands, and more. While these emails often contain valuable discounts and promo codes, they’re easy to miss or forget when we actually need them.

CouponAI solves this problem by allowing users to automatically forward promotional emails to a custom inbound server email address for parsing (powered by Postmark). From there, the parsed emails are processed with the help of a local LLM (Large Language Model) to extract structured information like:

• The company or brand
• Offer
• Offer description
• Expiry date
• Category (e.g., food, travel, shopping)

This data is then stored in a MongoDB database and can be retrieved via a simple frontend interface, helping users quickly check which discounts are available to them before making purchases or placing orders.

You can now retrieve all usable offers from a single endpoint and never miss a discount again!!

Demo

You can watch CouponAI demo here

As this project depends on a locally hosted Large Language Model (LLM) to process parsed email data — which would be costly to deploy in the cloud — it is currently not live.

However, you can run and test it on your local machine by following a few simple steps outlined below:

How to Run and Test Locally

You can try this project on your local machine by following these steps:

1. Clone the Repository and Install dependencies
   

   git clone https://github.com/Utkarsh212/CouponAI.git
   cd CouponAI
   npm install
   

2. Install and Run Ollama

   • You can download it from here
   • I have used gemma3:4b you can choose any suitable model as per your needs.

   ollama run <YOUR_MODEL_NAME>
   

3. Configure Environment Variables
   Update config.env with the following variables:
   

   PORT=5000
   MODEL=YOUR_MODEL_NAME // Add your model name here
   HOSTED_MODEL_URL=http://127.0.0.1:11434/api/chat
   DB_CONNECTION_URL=YOUR_DB_CONNECTION_STRING // Add your MongoDB connection string here
   

4. Expose the App with ngrok

   • You can download Ngrok from here
   • Run your Express app (e.g., on port 5000)
   • Start ngrok to expose it:

   ngrok http 5000
   

5. Configure Ngrok URL over Postmark

   • Add ngrok URL to your server’s inbound webhook URL field on Postmark portal.

6. Forward a Test Email

   • From any email account, forward promotional email to the Postmark inbound server email address.
   • Postmark will parse it and send the content to your server’s inbound webhook URL.

7. See Results

   • Open the index.html file to view the results.
   • You'll see a list of all promotional emails along with their details.
   • Alternatively, you can access the Ngrok URL directly to view all the extracted offers in JOSN.

Results

Here are example of an email that were forwarded to the inbound server:

Corresponding Output from LLM:

The parsed and structured JSON results:

{
  "company": "Zomato",
  "offer": "Up to 50% off on your favourite treats",
  "details": "Just code ZOMATO while ordering online.",
  "expiry": null,
  "category": "Food"
}

Code Repository

GitHub: github.com/Utkarsh212/CouponAI

How I Built It

The application is powered by a Node.js + Express backend, a locally running LLM model via Ollama, and Postmark’s inbound email parsing feature to automate the extraction of useful promotional data from email content.

Tech Stack

• Backend: Node.js + Express
• LLM: gemma3:4b model running locally via Ollama
• Database: MongoDB
• Email Parsing: Postmark Inbound Stream
• Development Tools: Ngrok (for exposing localhost), dotenv, axios

Architecture Overview

Below is the architecture diagram that illustrates the complete flow — from the moment a email is forwarded, to how it gets parsed, processed by an LLM, and stored in the database.

Flow & Architecture

1. Inbound Email Forwarding:
   Users set up forwarding from their personal email accounts to a Postmark-provided inbound email address.

2. Postmark Webhook Integration:
   Postmark parses the email and sends structured email data (subject, text body, HTML body, etc.) to a webhook on my Express server.

3. Email Parsing & LLM Processing:
   The server cleans the raw email content and generates a prompt, which is then sent to a locally running LLM (gemma3:4b). The LLM returns result which is then parsed into a JSON object with the following fields:

-   `company`: Brand or service name
-   `offer`: Promo description
-   `details`: Additional details about the offer
-   `expiry`: Validity date if mentioned
-   `category`: Context (e.g., food, shopping)

1. Storage & Retrieval:
   The structured offer data is saved in a MongoDB collection and can be retrieved via a public API route.

2. Frontend (Optional):
   For simplicity, I created a minimal HTML page that displays the available coupons pulled from the backend API.

Experience with Postmark

This was my first time working with Postmark’s inbound email stream feature — and it was a smooth experience. Setting up the inbound stream, getting the parsing to work, and configuring the webhook was very straightforward. I was especially impressed with how detailed and clean the parsed JSON format is out-of-the-box.