-
Example of slide from topic "Theory of Gravity""
Inspiration
Chat-GPT has the ability to create any form of text you can desire so how could we use this to automate our live?
What it does
Our program is a web based hacked where the front end asks for a topic, specific links about your topic, and number of slides. It then formats this information into queries fed into a GPT-3 model to generate content such as a summary of the links and bullet points of general information. It also uses Google Cloud's image search API to attach relevant images to the presentation. All of the content is then formatted into a presentation format to meet the requirements.
How we built it
For the front-end aspect of the hack, a simple html document was created with embedded Javascript to save the user's inputs. Simultaneously, a python script is running in the background to check for any new requests using Flask to read for any http requests for JSON files. For the back-end aspect, the links are first scraped into their raw text information to remove any unnecessary content. The information is then fed into the GPT model to summarize the content into bullet points. To find images, Google Cloud's Custom Search API is used to generate multiple relevant images, which are then selected to be included in the presentation. The presentation itself is generated through a python library which allows for the automated creation and formatting of powerpoint files through code.
Challenges we ran into
Some of the major issues we ran into were integration difficulties between merging the front and back-end aspects of the hack. Also, formatting the slides and parsed data proved to be very tedious and easily changed. Dealing with API tokens and query limits also proved to be a limiting factor to our testing. Of course, like any other project there was a lot of time spent on debugging and unit-testing of the code.
Accomplishments that we're proud of
This was the first time we both approached a project centered around API's and integrating AI models. We essentially had to learn everything involved, but we were very proud the results that were produced by our program.
Log in or sign up for Devpost to join the conversation.