You can one-hot encode categorical data like strings to build one big model. But if the data is independent, or if the tasks are orthogonal then you might be better served if you create two models.