Idee

Discover your next coding side project with the help of AI RNN text generation. Great for first year university students or those who are just looking for ideas to implement!

Comment

Inspiration

Seeing AIs create Guardian articles made us wonder what other types of creative problems an AI could help with. At the same time, we've all wanted to work on a side project at one point but were unsure of what topic to tackle. That's where Idee helps us out!

What it does

Idee generates unique coding project ideas. Just click a button on our app to get a unique AI-generated coding project name and description to give you inspiration for your next project.

How we built it

  • Created a GitHub scraper to collect the names and descriptions of trending projects
  • Ideally, we train a Machine Learning model (Recurrent Neural Net) with the GitHub project names and descriptions as the dataset
  • Then, use the above model to generate new text
  • But we found the training on our PCs was too slow, so instead we feed a GPT-2 API partial sentences to get it to generate coding project ideas using the GitHub data as context
  • Encapsulate the generation of coding projects with a Python Flask API
  • In the future, create an app frontend to call the API and show the generated coding projects

Challenges we ran into

  • Took a while to settle on an idea
  • We tried training RNN models on our computers, but it took way too long to produce a model that created coherent text.
  • We weren't able to find any services online that offered text generation ML solutions where we could train models using our data
  • GPT-3 would be a fantastic pre-trained model to generate text, but using its API requires being put on a waitlist
  • The GPT-2 API generates incoherent text sometimes and it loses the context of a coding project when we ask it to generate text for another part of the project (e.g. inspiration, what it does, how it works, etc).

Accomplishments that we're proud of

  • Learning about ML and RNNs
  • Generating awesome Figma designs
  • Creating a reusable Github scraper
  • Creating an API

What we learned

  • Learned about Machine Learning and RNNs
  • Learned about the uses of GPT-2 and GPT-3
  • Learned how to quickly put up an API using Python and Flask
  • Learned about calling APIs
  • Getting better at time management

What's next for Coding Project Generator

  • Train RNN models using our GitHub data to improve text generation accuracy
  • Figure out a way to ask an AI multiple questions about a single coding project
  • Implement the Figma designs with a frontend

Built With

Share this project:

Updates