What's NXT

Eliminate Errors Elevate Efficiency

Comment

Inspiration

Aneesh was texting someone and telling us about how autocorrect kept messing up his messages. We all agreed then started talking about our issues on how autocorrect messes with our lives. After doing some research, we realized how big of a problem this actually is and how many industries are affected by poor communication. Rhugaved then thought we should make a Grammarly but also implement full sentence generation with the power of AI. While developing our program, we also found it annoying scheduling or missing appointments, events, or deadlines, so we also decided to implement setting up instant appointments with Google Calendar so you can stay ahead of your schedule with minimal effort.

What it does

Smart AutoComplete: This feature suggests words or phrases to complete as you type, based on your previous typing patterns and context. This helps to speed up your typing and reduce errors, as you don't have to type out every word manually.

Auto Response Prompts: This feature provides you with suggested responses to common questions or messages. These prompts are based on pre-set templates or previous communication history. This can help to save time and ensure consistency in your messaging.

Auto-correction and Word Completion: This feature automatically corrects spelling mistakes and suggests words to complete as you type, based on a dictionary or machine learning algorithm. This helps to ensure that your messages are error-free and concise, without the need for manual editing.

Generate Calendar Invites based on messages suggesting meeting: This feature automatically generates a calendar invite based on the content of your message, if it suggests a meeting. This helps to streamline the scheduling process and ensure that all necessary details are included in the invite.

How we built it

Our extension uses GPT-3.5 turbo and various filters to analyze the chat history and current content of a typed message and suggest completions for the message. GPT-3.5 is a cutting-edge language model developed by OpenAI that uses deep learning to generate natural language text. The filters we're using likely involve adjusting parameters such as temperature, top-k sampling, and repetition penalty to fine-tune the output of the model.

Similarly to the Smart AutoComplete feature, our extension uses GPT-3.5 to generate response prompts based on the context of previous chat messages. This can be a powerful tool for saving time and improving the efficiency of communication.

Our extension uses an ML model that leverages n-grams and minimum edit distance to suggest word completions and correct spelling errors. N-grams are a type of language model that predicts the probability of a given word given the context of the surrounding words. Minimum edit distance is a measure of how many character edits are needed to transform one word into another. These techniques can help improve the accuracy and speed of typing.

Our extension also includes a feature that generates calendar invites based on messages suggesting a meeting at a specific time. This likely involves parsing the message using the modified timefhuman library to extract the relevant date and time information and then generating a calendar invite using a separate API.

In terms of software features, our extension is designed to work as a web extension on social media websites, extracting data and communicating with a Flask web server. The server hosts the word completion model, calendar links model, and OpenAI API, allowing the extension to access these resources as needed. Overall, our extension appears to be a powerful and innovative tool for improving typing and communication on websites.

Challenges we ran into

When we initially started working on the project, we had limited knowledge about language models. We began by training GPT2 for sentence completion, but later pivoted to using ChatGPT instead. The word completion model required us to incorporate a range of different words and expressions commonly used in social media. However, we encountered challenges with the frontend development of the extension, particularly in identifying the right classes and designing a user-friendly interface. Furthermore, the recent changes to the WhatsApp codebase created additional obstacles in our development process.

Another challenge we faced was in building the Flask server. We encountered issues with communication between the extension and the server, which required us to troubleshoot and refine our implementation. Despite these challenges, we remained committed to delivering a high-quality product that provides valuable functionality for users.

Accomplishments that we're proud of

Our team is extremely proud of what we accomplished in this 24 hour window. Although the product is only in its prototype stage right now we believe it shows enormous potential to assist everyday users in improving their communication by helping them with word suggestions, sentence completion, while simultaneously matching the tone of the suggestions based on the previous chats of the user with that specific person. We're also proud about the versatility of this product and how it can be applied in multiple industries and professional settings - ranging from customer service to emails with the boss, and in all of these situations will provide you with tools and assistance to strengthen your communication and connection with the other person.

What we learned

Through this project, we gained valuable experience working with the GPT-3.5 API and developing various models using text inputs. We also learned how to customize GPT-3.5's functionalities to suit our needs, such as adjusting tone, sarcasm, and sentence length. Our team developed expertise in integrating AI services with different platforms, including WhatsApp, Facebook, and Instagram. We also developed a deep understanding of n-grams and minimum edit distance algorithms and how they can be used to improve typing and communication. In addition, we customized an existing library to create the calendar invites feature, which gave us valuable experience in adapting existing tools to meet our needs. We experimented with different servers but ultimately chose Flask because it was the best fit for our requirements. Some team members also gained experience in building a web extension from scratch. Finally, we gained insight into the high demand for communication assistance tools, as evidenced by the popularity of tools like Grammarly. Overall, this project allowed us to develop a wide range of technical skills and gain a deeper understanding of the challenges and opportunities in the field of natural language processing.

What's next for What's NXT

In the future the team envisions for what's next is built around three collar pillars. 1- The web extension for our product will be published on the chrome extension store giving the world access to our product and all the benefits it entails. 2- The product will be expanded to other messaging services to expand its reach to as many users as possible. 3- In the same vein of expanding to as many users as possible the team wants the product to be able to function with as many different languages as possible so that we can expand our reach across all borders.

Share this project:

Updates