Code Vert

Code Vert!

Inspiration 🧠

Amidst the frantic rush to submit a semi-functioning project to a hackathon, it can be difficult to take time and consider the possible repercussions of your project on the world if it were to become a reality. And that tends to be exactly the kind of foresight and circumspection judging staff looks for in a prospective winner.

We wanted to build an application to save hackathon participants (1) assess the environmental risks of their projects and (2) save some time on creating their presentations.

What it does 💪🏽

Code Vert has two key features. The first is environmental risk assessment and mitigation strategies based on the Devpost you have written for your project. The second is script and image asset generation for your final project presentation, designed to save you some time and give you some inspiration about how to best present the fruit of your toil.

How we built it ⚙️

Code Vert is written in Python and deployed on Streamlit. I uses the OpenAI API to generate text and images with DALL-E and ChatGPT (for the script/environmental analysis and image asset generation, respectively).

Challenges we ran into ⛰️

One major challenge was getting the OpenAI API to respond in JSON for the environmental analysis. After a lot of tinkering, I came up with a reliable prompt that worked on multiple consecutive generations.

Accomplishments that we're proud of 👑

Solving the JSON issue is definitely something we're proud and it's a technique that we know will come in handy again in the future. We're also proud of working with the OpenAI DALL-E text-to-image endpoint for the first time. Finally, we're proud that we were able to deploy our project to the Streamlit Community Cloud.

What we learned 📈

We learned about the power of generative AI in enhancing human creativity and the power of Streamlit in facilitating quick, seamless online deployment of data-based web projects.

What's next for Code-Vert 🚀

The image assets still come out looking a little wonky, so one approach I'd like to try is removing the stop words from the presentation script before submitting the script lines as descriptions for the images. This would use the NLTK library (Natural Language Toolkit) and I think the images would come out more coherent and connected to the script.