Video game character chatbot with the OpenAI API

This video game character chatbot started in early 2023, just as the OpenAI API was opening up, as part of a design course (UML and databases). The idea: a user picks a video game, then a character, and opens a conversation with them.

It began as an Express app with server-rendered views, before being rebuilt as a Node.js API consumed by a React front end.

For the character to stay believable, the model has to know their universe and remember what was already said. But the API keeps no state between calls.

The database also had to be clean and match the course’s conceptual model, linking users, games, characters, conversations and messages, all behind authentication.

When a character is created, the app asks OpenAI for a detailed description, which then serves as the system prompt. With each new message, the whole conversation history is sent back to the model, which answers in character.

I designed the MySQL database from the logical data model (with an SQL view for some queries), wrote the CRUD for users, games, characters, conversations and messages, and secured the API with JWT and bcrypt-hashed passwords. The React front end offers a debounced game search, a cover gallery built with Swiper and the chat page. Mocha tests cover part of the routes.