We use XGBoost as the core learner for reinforcement learning at https://improve.ai despite the popularity of neural networks in academia.

With tabular or nested data a human has already done a lot of work to organize that data in a machine friendly form - much of the feature engineering is performed by the data schema itself.