GameRecommend AI

Inspiration

Finding new videogames to play is exhausting, as it can take a long time to choose a new game to try, only to find out that it isn't your type. Many game recommenders out there utilize filters, and thus can only work if the game is in the database. On top of that, using filters can end up missing many aspects of a game.

What it does

GameRecommend AI utilizes AI to find similar games to any game you like. All you need to do is enter in a game you like, press the button, and the AI will return 5 games it thinks you will like, along with reasons why.

How I built it

The recommendation system uses GPT-4o mini from OpenAI API. System instructions are given to the model, including specific JSON formatting, to not make up games, to use the games' official names, and to give exactly 5 reasons why the games match.

Then, IGDB API is called to get the cover art. The title of each recommended game is sent in different API calls as a search, and it returns the image ID of the cover art. Because there's only 4 API requests per second allowed, and 5 games, the loading is artificially slowed down to allow enough time for all of the images to be called.

The information from OpenAI API and IGDB API are then styled and displayed, along with a simple link to a Google search using the title name.

Challenges I ran into

The short timeframe was very challenging, but it was definitely a great experience. Fitting in the idea-making, the OpenAI API calls, the React programming, the IGDB API calls, the formatting of the IGDB API responses, the logo design, the written submission, and GitHub readme all into a couple of days without a team was very difficult. I was able to lessen my workload a bit by using a Vercel template to set up the project, and a GitHub readme template.

Accomplishments that I'm proud of

I'm proud of committing to finishing the project before the deadline. At many points, I felt that I wouldn't have the time to complete the project and I should call it a day. However, this is my first experience in a weekend-long hackathon, and I wanted to gain that experience of working under a time pressure. While the project is pretty simple, I'm still very happy with it.

What I learned

While I've used React and Next.js in the past, I definitely got a much deeper understanding of the tools by making this project. I also got more practice in API calling and Tailwind CSS.

What's next for GameRecommend AI

I definitely think the IGDB API images can be improved, as sometimes the title from GPT-4o mini doesn't find any games in the IGDB database. This is pretty rare, but a slightly more frequent bug is if the title is very similar to the title of another game, it may get the cover for the other game instead.

Additionally, right now the recommendation from GPT-4o mini is only taking in the title of the game. For very small indie games or very new games where there isn't any information about it in OpenAI's database, GPT-4o mini is guessing what the game is about from the title. Because IGDB gets information from new games on Steam very quickly, getting a description of the game and passing it to GPT-4o mini can make the results much more accurate for very small indie games or very new games.

Homescreen Home Page