DEV Community

Cover image for InboxOps – AI-Powered Email Automation Platform Using Postmark Inbound Parsing

InboxOps – AI-Powered Email Automation Platform Using Postmark Inbound Parsing

Mihir Amin on June 07, 2025

This is a submission for the Postmark Challenge: Inbox Innovators. What I Built InboxOps is a powerful AI-augmented operations platform...
Collapse
 
axrisi profile image
Nikoloz Turazashvili (@axrisi)

FYI the link to demo is not working: inboxops.render.com

Collapse
 
dynamicmortal profile image
Mihir Amin

The link had a silly typo i didn't knew it was .onrender.com.
The Submission is updated with Link!

Collapse
 
axrisi profile image
Nikoloz Turazashvili (@axrisi)

good :) now I can access it. Any way I can test it? maybe sending email with some fake employer request?

Thread Thread
 
dynamicmortal profile image
Mihir Amin

Yes, the Email address is:

10bb06248998326c0167cda19c82da62@inbound.postmarkapp.com
Enter fullscreen mode Exit fullscreen mode

The schema to send an Employee or HR Leave application is suggested:
Subject: xyzabc
Body: Start Date, End Date
The body can contain anything!

Thread Thread
 
axrisi profile image
Nikoloz Turazashvili (@axrisi)

thanks! jsut sent email and saw it in inbox, but not in other parts of dashboard, should I make some action to trigger it go through the flow?

Thread Thread
 
dynamicmortal profile image
Mihir Amin

Yes, the Ticket was not getting created, the whole workflow was at a standstill due to Postmark not recognizing the email address.
I made some changes in the backend routing code, it should work just fine now!

Collapse
 
dotallio profile image
Dotallio

Super impressive how you’ve combined Postmark parsing with LLMs to automate and unify so many workflows. Curious, how did you approach prompt chaining for reliable intent and summary extraction across really different email types?

Collapse
 
dynamicmortal profile image
Mihir Amin

Yes when it came to extracting intent, summary, tags, and criticality from different email types Orders, HR, Customer Enquiries support, Approvals I used structured prompt chaining.
The Postmark Inbound Email JSON is passed to LLM to return in a structured JSON format:


{
"intent": "support",
"summary": "Customer reports duplicate charge and requests refund.",
"criticality": "high",
"tags": ["refund", "billing", "duplicate"]
}

This solved me the consistent storage of data. Entirely the workflows cannot just rely on LLM. To improve reliability across types: Order Emails, Approvals, HR, Customer Enquiries Support Requests Issues Tickets, I did:

  • i made use of few-shot examples in prompts of Google Gemini LLM.
  • Fallbacked to RegEx for edge cases
  • Added prompt to generate auto-replies based on the Sender Mail! The Postmark Inbound Mail processing JSON is the major part!
Collapse
 
nathan_tarbert profile image
Nathan Tarbert

been cool seeing steady progress - it adds up. you think the roadmap or just showing up every day keeps stuff moving the most?