replies.ai

The Wihte-Mailers!
Registration Page
Google GmailOAuth Log on
Shows unanswered emails
Generates suggestions/templates

Inspiration

Communication has been an integral part of human survival for the longest time. This world completely changed with the introduction of a new tool called electronic mails. Humans had started working towards fast and efficient communication. Another step to this race has been the introduction of AI. Leveraging exactly this, we present replies.ai.

What it does

replies.ai is a powerful tool that accumulates all your unanswered emails at one place, not only that, it gives you personalized suggestions and draft reply templates. Now you can use our AI-powered suggestion and send it automatically, edit it yourself or regenerate a new template!

How we built it

Our web application leverages the React framework supplemented by Google OAuth and Gmail APIs alongside Gemini APIs. When the user onboards using OAuth, their emails from the last 60 days are parsed through the LLM which picks the top 5 referencable emails and divides them between 3 formality categories, including Casual, Professional, and Formal. When the user opens an unanswered email on the application, the email's body is parsed through the LLM again to write a suggested reply based on the sentence structure and vocabulary referencing the previously sent emails.

Challenges we ran into

Creating dynamic end-to-end projects comes with a variety of challenges and endless debugging. We are proud to say that our team handled bugs very efficiently and overcame challenges as a team in collaboration. One of the major challenges we ran into was with connecting OAuth authentication to Gmail authentication, as well as Firebase authentication. Initially, we researched dynamic and optimized alternatives, however, with the time constraint, we decided to remove the database functionality and store data on local servers to achieve our minimum viable product.

Accomplishments that we're proud of

With the changing world and new technical introductions every day, we are proud to keep up with the artificial intelligence phase of things. We believe we did a great job leveraging large language models specifically Gemini APIs, and put it to a use that has very high social impact. Our product can potentially save multiple hours and make tedious jobs easier. While spending the last 24 hours together, we also got time to learn more about each other and also our own selves. We started to build a great team and started finding motivation and happiness by accomplishing short-term goals.

What we learned

This weekend was not only an opportunity to improve on our soft skills but also gave a platform and a team to grow technically. The technology stacks that our team researched on and implemented are Gmail API, Firebase OAuth Authentication, Gemini API. Our team had very diverse technical backgrounds, which helped us share our experiences and increase our breadth of technical knowledge, as well as become better engineers.

What's next for replies.ai

replies.ai is an extremely powerful tool with very high potential to compete in the market. We aspire to scale this project and allow the user to train the model on their own by feeding it data in the form of articles, blogs, or manually written email templates, and fine tune the API calls for a better and more accurate response. We also rejected the idea of a chrome extension during the start of the project due to time constraints. We aim for an extension that pops up when the user opens their Gmail/Outlook and provides suggestions directly on the email provider's user interface. We also aspire to create our machine learning model and train it to better satisfy our need with increased efficiency and accuracy.