AI-U | Devpost

Inspiration

Amidst the arising scene of AI chatbots, our team aspired to utilize the chatbots power to make an impact for individuals who seek self-improvement and self-awareness. To do so, we were intrigued with producing AI generated versions of ourselves to seek ourselves from a different perspective.

What it does

AI-U is an interactive text-based website game where the user observes an AI simulation of themselves going through three scenarios: an interview, a chat with a friend, and a public speaking task. As a result, the user is able to identify any flaws or weaknesses among themselves within these scenarios in order to promote self-improvement. To create the AI simulation of themselves, the user feeds the game with their text messages to allow the model to analyze their speech patterns. From there, the AI aims to emulate its analysis.

How we built it

For the frontend tasks, we utilized Figma and JavaScript. For the backend, our team implemented a large language model using OpenAI's public API as well as Replicate's API to consume and analyze inputted text data given from the user, and retrieve an imitation of the users conversational tone. Using these APIs we were able to access the Large Language Models Chat GPT-3.5 and Llama 2.

Challenges we ran into

Our team found difficulty during the single-page application process as it was a learning curve to get started. Moreover, we found the research and cost aspects to make the model optimal challenging. We needed to find a model that was accessible, affordable, and least time-consuming. As well, a huge issue was figuring out ways to use our time efficiently when each component of our project relied on another. We had to make assumptions to complete our separate tasks, all the while allowing for flexibility in our implementation.

Accomplishments that we're proud of

We are proud of the completion status of AI-U considering the time constraint. Our original idea was able to come to life, and although this project is not flawless, we believe the end result was created to the best of our abilities. We are satisfied with the user interface application, backend management, and the model implementation.

What we learned

We learned the value of pre-planning and researching to encourage efficiency and certainty towards the project. In addition, we were able to learn new concepts such as Large Language Models and AI in general. It was also of value to learn when to try out new ideas instead of being hung up on something that would not work. This allowed us to progress at a good pace.

What's next for AI-U

AI-U strives to expand from its text-based implementation to incorporate 32D visual scenes and actions, providing AI-generated visuals within the game. Our team also hopes to pursue a more accurate model to accomplish a better replication of the user to observe and relate to, all within justified morals. As well, we would aim to allow for more user interaction and have the ability to adjust the AI-imitation while the game is running.