BOBUX BARGAINS

hello

Inspiration

Whenever we think about finance software, we think of boring, unintuitive software, with exorbitant licensing costs and little accessibility or ease to the general public to understand. Oftentimes, processes are also hard to visualize. Therefore, we want to combine the power of AI agents and common variety game engines to simulate a market.

What it does

Bobux bargains simulates a real world economy including supply and demand and financial transactions, on Roblox! There are two kinds of AI agents at play in our project, Buyers and Sellers. Buyers simulate buyers of goods and services in the real world who want to find the best price available for a product in a good and services market. Sellers are businesses that have the product buyers want and aim to earn the most profit out of each transaction. Each AI agent is fully customizable and can use a variety of AI models using the Open Router API. This allows the application to test the reasoning and negotiation capabilities of each model. Bobux Bargains is also scalable, allowing more accurate tests and economic modelling.

How we built it

First, our environment of choice is Roblox and Roblox Studio, as the games produced on the platform are highly accessible to anyone with a device. Furthermore, it is an easy platform to quickly prototype and develop 3D environments with player models, making it easy to attach our agents to an NPC. The agents print their conversations to the Roblox game NPC chat. For our actual agents, we are using OpenRouter to allow us to easily pick different LLMs to experiment with and potentially build agents with different personalities. To develop our prompts, our application keeps track of historical negociations that the buyer has conducted and feed them back to the language model, which means the buyer can remember deals it has made and make decisions based on past experience to conduct future deals.

Challenges we ran into

Our challenges actually mainly lies with building the simulation environment itself Getting automated Roblox characters to communicate with each other Getting elements of the game to actually interact with each other in the physical space in a way that makes sense Dealing with race conditions between element rendering and agent / other API calls within the game engine Version control

Accomplishments that we're proud of

First and foremost, we are really proud of our concept and design, which also scales to various other AI Agent simulations and tasks. We are also proud to have built a full-stack application within the time constraint and within such a unique environment.

What we learned

A key learning includes how to simulate negotiation. We often intuitively know how to conduct negotiations ourselves at least at a basic level, but trying to instruct the AI agents to conduct sensible negotiations requires a deeper understanding of the mechanics of sale negotiation. Furthermore, we also learned how to build games and work with 3D spaces effectively. For the most part, this our first part ever building in 3D environments, so from collision to movement, we now have a better understanding of 3D game engines.

What's next for BOBUX BARGAINS

We want to expand BOBUX BARGAINS to have more buyer and seller agents, while also allowing players to negotiate with other NPCs and players. With 10 BILLION agents, we can spend $450 000 PER SECOND in OpenRouter credits and simulate the BEST and MOST ACCURATE economic model of the real-world marketplace there is.