Inspiration
We realized that a lot of the time when we're trying to find specific information online, it can be hard to get the browser to show exactly what you're after, and this is coming from a pair of digital natives who have been used to browsers our whole life. It must be much worse for those not used to the internet.
What it does
We use the openAI API to take what the user wants to find out and turn it into a searchable query optimized for a search engine of choice.
How we built it
The openAI API is accessed through python using flask and HTML to host a server locally. The flask initiatives a function using the user's input to query gpt 3.5 on what the best way to formulate the input into a search query.
Challenges we ran into
Hiding the API key was a challenge as we don't want to publicly show it. We were originally planning on hosting the website as a static page from our github repository, but from what we gather that's impossible without having the API key be public. Instead we create a .env file with the API key, and don't post that file to the repository, rather requiring would be users to get an API key themselves.
Accomplishments that we're proud of
We're proud of integrating use of this API into a working web-interface. For Lars, this is the first time working with APIs at all so it's really cool seeing how seamlessly it's possible to make them work.
What we learned
Lars learned a lot about using APIs and particularly the openAI API, while Opemipo learned about using flask as a web development tool. We both learned that working together in a team is much more fun and efficient than working alone.
What's next for Search-GPT
We also wanted to make a feature where the model comes up with a list of queries, performs each of them and evaluates them, only to return the best one to the user, however to do this we need to use gpt-4 which we couldn't get to work with out API key, and it would take too much time. This means that implementing this would be the logical next step. If this works really well, we could continue to set up a proxy backend server that has the api key, and sends results to a frontend server which is available for users, making search-GPT actually available to the public.
Log in or sign up for Devpost to join the conversation.