-
Inbox Page
-
Documents
-
Sources
Inspiration
As organizers of HackMIT, our team handles thousands of emails over the semester leading up to our big event! In order to respond to help emails more efficiently and boost our team's productivity, we created Pigeon!
What it does
Pigeon automatically tracks help emails by forwarding emails sent to our mail server to our custom made client. By uploading documents about our event as source text to be used in our generated response, we query the corpus using semantic search from a Redis vector database of document embeddings, and then generate an email response using a LLM. You can then individually edit and send the email within our client and track the documents used to generate the response!
How we built it
Our backend was built with flask and our frontend with Typescript. We used a SQL database to store emails, a Redis vector database to store documents and embeddings.
Challenges we ran into
Some of the challenges we ran into include incorporating the Redis stack and ensuring our emails were received and sent reliably. Additionally, we had to generate a large enough corpus of documents to ensure our responses were accurate.
What's next for Pigeon: Your Personal Email Assistant
We would love to fine-tune our prompt better to ensure the LLM utilizes the documents even more effectively. Additionally deploying our app to the cloud to make it universally accessible.
Log in or sign up for Devpost to join the conversation.