Inspiration

The project idea first came to mind based off "AI Family Guy", a 24/7 livestream of AI generated Family Guy content. We extended this idea to the world as a whole.

What it does

Sims.AI simulates the world, relationships, and emotions, streaming this data in real time to a 3D experience for visualization.

How we built it

We used Unity for the frontend and FastAPI + Gemini on the backend for AI simulation and state management. Models were made in Blender or generated by Meshy.ai.

Challenges we ran into

Streaming data in real time to Unity and using this data to affect a 3D experience is a major challenge. We had to really optimize our operations in order to reduce latency as much as possible. Creating the world environment itself was also a notable difficulty due to our desire to create a polished final product artistically AND functionally.

Accomplishments that we're proud of

The system we implemented on the backend was developed independent of any research papers, but after further digging is in reality an enhancement of Stanford's 2023 research paper using LLMs to simulate social interactions. Developing a robust system that emulates the capabilities of code written by PHD researchers is a feat that we are very proud of.

What we learned

Using asset packs for 3D models is a powerful tool that should be done whenever possible to save time. Modelling is an incredibly time consuming process that really ate away at the time we had.

What's next for Sims.AI

We hope to implement more actions, spaces, and expand the world in terms of size to allow for greater variety in the interactions that take place.

Built With

Share this project:

Updates