How to reduce interaction for training model?

Question

After reading this topic https://learn.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/load-data-model-builder it seems to me that training a model is a rather manual task and I'm not talking about data preparing. You have to manually select types, data files to load and so on. Is it possible to make it more code driven? For example, instead of all those manual actions can I call some kind of model training API, specify all necessary types and then feed my data in as a stream? Feeding data as a stream is especially important part. I'm thinking about dynamically preparing and feeding the data. Technically I can prepare the data and save it as a file and then use it but ideally I would like to skip the step of writing the data file first.