Unprofitable Startup

Some of the core features of this project were:

• PyTorch Neural Network is used to predict a company's net income. This uses publicly available financial statement data as inputs to create then and train a neural network model capable of forecasting a company’s net income. This process also included different metrics used to analyse and determine the best architecture for the neural network.
• Integration of a neural network model with a game engine. Using ONNX file format, an intermediate file format representing a neural network, and the Unreal Engine Neural Network Inference plugin (NNI), the neural network can be run in the game engine. Besides the model representation, this also required handling exporting information to properly normalise data to the 0 to 1 range used by the neural network. In the Unprofitable Startup use case, the neural network simulates the company’s performance at the end of each month.
• Business simulation game mechanics. As the core gameplay mechanics, the player can allocate the budget of the company he is managing, as well as see in detail the different metrics of the company. The player’s actions, in this case, affect the inputs and, therefore, the output produced whenever the neural network simulation is run.
• Usability features to simplify the gameplay. Strategy games can be complicated and challenging to understand for players. Multiple usability features were implemented to alleviate this issue. Like rounding financial values to millions or hundreds of thousands, making sliders to allocate the budget snap to determined step amounts, a help section explaining how to play the game, and a UI highlighting the company's current state.