Worked on enhancing broadcasting robustness in the jax-ml/jax repository by developing a feature that enables reliable handling of mixed scalar and tensor inputs, including support for bfloat16. This involved updating type checking in the _bcast path and converting scalar inputs to MLIR constants, ensuring consistent behavior across numeric types. In the ROCm/jax repository, addressed a bug related to the conversion of Python native types and NumPy scalars to MLIR constants during MLIR conversion, improving reliability when broadcasting mixed data types. The work leveraged Python and focused on compiler development, numerical computing, and machine learning infrastructure within the JAX ecosystem.