Mail Dolphin Documentation

Team Members:

• Mir Muhammad Abidul Haq (Ahnaf) - email: [email protected]
• A. N. M. Noor - email: [email protected]
• Iqbal Samin Prithul - email: [email protected]

Overview of our Repository

• static - this folder contains the codes and design elements for the client side.
• templates - this folder contains HTML template files.
• app.py - This is the main code for flask, tinyllama, and the API integrations.
• requirements.txt - Contains a list of required packages fo this project.

About our Project

Problem Statement

Managing and responding to emails can be time-consuming and overwhelming, especially when dealing with a high volume of incoming messages. It is often necessary to respond to emails promptly and in a personalized manner to maintain professional communication and ensure that important information is communicated in a timely manner. As a result, it is more difficult to understand the intent behind each email and respond appropriately based on individual preferences and contexts.

Solution Overview

To address this issue, we've developed an automated email reply system (Mail Dolphin) that leverages advanced natural language processing (NLP) and machine learning techniques. This system is designed to: Analyze Incoming Emails: The system scans and processes incoming emails to extract key information, such as sender details, subject lines, and the body of the email. It uses NLP to parse the content of the emails and identify relevant keywords, phrases, and contexts. Craft Personalized Replies Based on Your Preferences: The system generates customized replies that align with the user's communication style and preferences. It uses predefined templates and dynamically fills in the details based on the context of the incoming email. To ensure that the recipient feels valued and understood, the replies are carefully crafted to be professional, polite, and contextually appropriate.

Workflow

1. User integrates their existing Gmail account with Mail-Dolphin.
2. Adds some General information along with some example emails and replies to tune the service to match their preferences.
3. Mail-Dolphin checks for any email the user receives.
4. Generates a reply based on the information and examples given by the user.
5. Replies to the email on behalf of the user.

Tech Stacks

We used,

• Flask API for the backend.
• Bootstrap, CSS, and HTML for the interface.
• Javascript for handling the communication between the frontend and the backend.
• Tinyllama for Natural Language Processing.
• Gmail API to integrate user's Gmail with our Mail-Dolphin.

Starting the project on your local machine

1. Install python (3.10 or higher) and clone this repository.

git clone https://github.com/Ahnaf-nub/Mail_Dolphin-by-hardware_synapse.git

2. Head into the directory.

cd Mail_Dolphin-by-hardware_synapse-main

4. Install required packages.

pip install -r requirements.txt

5. Run app.py

python3 app.py

Usage

Customization

Tuning Mail-Dolphin is really easy!

First, add some general information about yourself / your company. Anything that would be needed for Mail-Dolphin in order to reply to emails on your behalf. Go to the Tune page. Here, you'll find the customization wizard. Now add to the General Information.

Then, add some example replies that describe your desired tone. It is a good idea to add some examples based on frequently asked questions. Inside the Example replies section, you can add, delete, or duplicate an example.

More Customization

You can also filter emails based on email addresses and intents. Head over to the Filtering section, and make some changes!

Here, Mail-Dolphin is not to send replies to [email protected], and [email protected]. Also, Mail-Dolphin is to send replies If only someone wants information about our service.

After you are done, click the save icon to save your tuning info.

Testing

You can experiment by sending some demo emails to our service. Head over to the Test page and send a test email to the server. Mail-Dolphin would generate a reply based on your settings and show it to you. Do more tuning if the reply is not satisfactory.