This is a submission for the Postmark Challenge: Inbox Innovators.
What I Built
I have build a tool "InboxMemory AI". In short its think of it as ChatGPT for your emails. You can send emails to make it into searchable knowledge base and ask questions via email or from the app in natural language. Basically RAG from emails
Demo
Video Walthrough
screenshots
QA via Email
QA via App
Getting Started
Just send "Hi [your name]" to the email [email protected]. The AI picks up your name, creates your account, and you're ready to go ⚡
Behind the Scenes
When you forward an email, we break it into chunks and store it in our vector database. If you send a question, our AI searches through everything you've saved and gives you an actual answer. You can also hop into the web app to browse your saved emails or ask questions with a proper interface.
Code Repository
GitHub Link
ashiqsultan
/
inbox_memory_ai
Inbox Memory AI build for Postmark hackathon | Using AI via email
InboxMemory AI
📁 For detailed setup and development instructions, please check the respective READMEs inside the
backend/andfrontend/folders.
Postmark Hackathon Project
This project was built for the Postmark Challenge: Inbox Innovators hackathon.
🔗 Blog Post: Read more about this project
What is InboxMemory AI?
InboxMemory AI is like ChatGPT for your emails. It's an intelligent email assistant that transforms your inbox into a searchable knowledge base using RAG (Retrieval Augmented Generation) technology.
Key Features
- 📧 Email-to-Knowledge: Forward emails to create a searchable knowledge base
- 🤖 Natural Language Q&A: Ask questions about your emails in plain English
- ✉️ Email Interface: Query via email or through the web app
- 🔍 Smart Search: AI-powered semantic search through your email content
-
⚡ Quick Setup: Just send "Hi [your name]" to
[email protected]to get started
How It Works
- Store: Forward emails to build your knowledge…
How I Built It
Techstack
- FastAPI for the backend
- PostgreSQL for email storage
- LanceDB as our vector database
- Redis for OTPs and rate limiting
- React for the frontend
- Gemini handling the AI magic and embeddings
- Docker for deploymnet
Along with the RAG system I ahve also build Auth system with email OTP because why not? since Postmark an email service anyways why not lets try that out. Gemini provides generous free tire which allowed me to use them. Wrote smaller AI Agent like files with specific system instructions for each task.
Some challenges I faced
I initially used Ngrok when testing in local but suprised to find no ngrok like services was mentioned in their docs. It will be a pain to initially debug without them
Architecture
Please try it
Please try it out as of writing this I have 50 more free emails from Postmark gota use them all. If this sounds useful to you, an upvote would be great. Thanks.
Top comments (10)
pretty cool seeing you get this working end-to-end tbh - always blows my mind thinking about how much stuff just sits in email, unused. you think most people even realize how much value is hiding in their own inbox?
Awesome work, @ashiqsultan 🔥 I think Gemini for Gmail does something similar too—curious how yours is different!
Well, I haven't tried Gemini in Gmail, but yeah thought of building something cool for the hackathon and ended with with this idea. 🙂
Love the idea of making email knowledge actually searchable and usable with RAG - super practical. Are there any surprising use cases you've seen so far?
Great project 🔥
Thanks for checking out.
Nice idea! Great work!
Very Informative
Nice!
hulk think tool good! hulk once smash mouse in lab, mouse still work. hulk email mouse incident to self. now hulk always remember funny mouse story. hulk try "inboxmemory ai" next time. keep up good work! 💪