Inspiration
The inspiration for this project was the overuse of AI, especially for small tasks. In addition, the high cost of APIs and the carbon footprint left by LLMs made us feel compelled to find a response to the issue.
What it does
The web application allows users to enter your OpenAI API to automatically switch between different GPT models depending on the complexity of the prompt using Nvidia's classification model. In addition, the application reveals to the user how much they are reducing their carbon footprint and saving money at the same time.
How we built it
The web application was built with a classic Flask backend and React frontend. The main model used to classify the prompt was the Nvidia prompt complexity model available on Hugging Face. In addition, chats are stored on a NoSQL database, Firestore.
Challenges we ran into
We had major issues adding user authentication into our product due to the nature of users having to enter their API key. In addition, our team was lacking in terms of front end development experience, which made that part of the process quite lengthy.
Accomplishments that we're proud of
The web application works great and functions just as planned.
What we learned
Better planning from the start could have made integration much easier in terms of user authentication and front-end features.
What's next for Green Prompt
User authentication, better design, and maybe the addition of other model API's and offline models.
Log in or sign up for Devpost to join the conversation.