Worked on enhancing data flow control and encoding robustness across the pytorch/tensordict and pytorch/rl repositories. Developed a feature in ProbabilisticTensorDictSequential that allows users to specify selective output keys, giving finer control over tensor dictionary outputs and improving usability for downstream modeling. Addressed a bug in pytorch/rl by correcting batch size handling during Composite encoding, ensuring accurate TensorDict creation. Expanded test coverage to validate output shapes and batch sizes, reducing the risk of silent errors. Utilized Python, data structures, and testing methodologies to deliver reliable, maintainable improvements that streamline data processing and support machine learning workflows in these libraries.