Carl AI

Inspiration

We noticed that many students do not have the opportunity, connections, or money to afford quality research programs and mentors. We wanted to create something that provided underprivileged students with personalized help and advice, like a paid research mentor would, at absolutely no cost.

What it does

Carl takes inputs given by the user and uses a trained OpenAI API model to give specific and thorough research information/advice back to the user. It uses an iMessage type of UI to give the feeling that the user is messaging their personal mentor.

How we built it

We built the website with WordPress. We used the WordPress code editor to edit the HTML and CSS of the website to create a more attractive look. For the AI, we used the AI Engine plugin on WordPress and input an OpenAI API key. We then trained the model with instructions and data through MeowApps.

Challenges we ran into

The OpenAI API key did not work at first so we tried different API keys like Gemini API from google and an Anthropic API key but neither of those worked properly either. After troubleshooting the OpenAI API we found that we needed to fix the billing address to the account and match it to GPT-3.5 Turbo 16k and it began to work. However, the responses and tone it was giving were not good enough to help properly aid as a research mentor. Therefore, we had to fine tune parameters on the backend. We lowered the temperature of the model to decrease its randomness and also adjusted its max tokens to change the length of Carl’s responses. This fine-tuning process trained Carl to be a very useful research assistant.

Accomplishments that we're proud of

We're proud that we were able to create a completely free and running website that anyone can use online. The website looks professional and attractive. We're also proud that despite roadblocks and challenges, we were able to add and train a unique AI on our website.

What we learned

We furthered our proficiency with website builders and learned important fundamental knowledge on the way websites function. We also got a lot more familiar with using APIs and getting them to work. Fine-tuning an AI model was also a great learning experience, and we are excited to find more ways to use AI for good in the future.