We were helping a fast-food giant develop automated order-taking technology that would enable novel drive-thru layouts, such as parallel order lanes or multiple pick-up windows. However, decision-making was frozen as we couldn’t predict the consequences within the kitchen from these concepts and the client didn’t possess a low-fidelity method of testing them.

So, I built an interactive drive-thru simulator. Cars drove in, ordered, and collected their meals when they were ready. I modeled traffic using the Intelligent Driver Model, so the cars would flow realistically. Time estimates for the order stations were based on prototype testing, and the orders, kitchen roles, and prep times were driven by the client's exhaustive data about their processes. All of this was custom-built in Python and Streamlit.

The client relied upon the web app for months, comparing estimated performance metrics for various configurations. The simulation particularly helped resolve disagreement around one contentious concept, which half the team felt would speed things up and half the team felt would slow things down. A year later, my colleagues ordered a meal from the finished product when it debuted in an actual restaurant.