You tried several models and tuned them appropriately and found the best performing model. What is the typical approach taken after this model selection step to improve the solution? E.g., is it feature engineering? If so, what are typical approaches for feature engineering? Please provide links to relevant resources.