What I learned building chatbots for banks and dealerships

For several years I worked at SOYBOT, a chatbot system that started out handling inquiries from people interested in buying cars from dealerships, and later expanded into other industries, including banking. I went from solving small problems as a trainee to leading a team of four, with chatbots deployed on more than 300 websites, three WhatsApp numbers, and twenty social media pages.

One of the first things I built was a C# (.NET 6) API that compared two vehicles and generated an image with the comparison. Later I built a system that parsed the dealerships’ huge spreadsheets —makes, models, years, specs— and converted them into an XML the chatbot could query to answer with real data during a conversation.

The most interesting part, technically, was bringing ClearScript into the bots’ scripting: it let us run JavaScript in real time from C#, running .NET libraries from that JavaScript. That flexibility was key to iterating quickly on conversation scripts without recompiling the whole system.

That year we competed at the 2019 Chatbot Awards and won the Startup category. It was one of the first times I felt that what we built day to day had a measurable impact outside the company.