Terminator

Mobile Screen
Desktop Screen 1
Desktop Screen 2
Desktop Screen 3

Inspiration

The inspiration for this project came from our vision for AI agents simulating environments for various applications, including medical training, corporate training, language learning, and many more. We wanted to use the power of Cerebras inference to build something that wasn't possible using other LLM providers. As inference speeds improve, more applications become viable and reliable. The faster the inference, the more use cases can be built on top of it. As LLMs continue to improve, we believe inference speed will be the key differentiator for future applications. That's why we wanted to bring our vision to life through a game.

What it does

This project features an AI overlord that has access to all inputs provided by the AI Agent Orchestration system. Multiple specialized AI agents - including a shooting agent, strategy agent, and score-tracking agent - work together to feed information to a Decision Agent that makes the final call to action, similar to human brain architecture. The game allows users to play a shooting-style game against the AI Overlord, creating an engaging battleground between human intuition and AI strategy.

How we built it

We spent considerable time on ideation, starting with an ambitious vision that was too large for the hackathon timeline. We had to distill our idea down to its core essence: building a mixture-of-agents model applied to a simulated environment. After finalizing our concept, we first built the game engine from scratch, then integrated the Cerebras APIs for all the agents. We also developed a mobile UI because we prioritized user convenience - playing games on phones is more intuitive than on laptops. Throughout development, we kept the end user experience at the forefront of our decisions.

Challenges we ran into

Building a game engine from scratch was our primary challenge. Having never done this before, we didn't anticipate how time-consuming it would be. Eventually, we successfully created a scaled-down version of our original vision. Our second major challenge was integrating multiple agents into a coherent UI that showcases their thinking processes and outputs. Without existing examples of applications displaying multiple AI agents debating or revealing their reasoning, we had to innovate and create our own visualization approach.

Accomplishments that we're proud of

Building our own game engine from scratch - an incredibly challenging but rewarding achievement Creating an AI agent orchestration system using Cerebras inference with near real-time performance Implementing multiplayer capabilities Developing a comprehensive web interface where viewers can watch two players simultaneously while observing the AI overlord's thinking process and all supporting agents' reasoning

What we learned

hrough this project, we gained valuable experience in:

Building a game engine from the ground up Streaming gameplay to multiple devices simultaneously Creating a unified interface to display parallel gameplay sessions Working with ultra-fast inference using Cerebras Orchestrating multiple AI agents effectively

All these skills were completely new to us, making this an incredibly rewarding journey.