During July 2026, contributed to the jax-ml/jax repository by implementing support for negative axis indexing in the parallel primitives lax.psum_scatter and lax.all_to_all. This work leveraged the existing canonicalize_axis utility to normalize axis arguments, ensuring consistent behavior for both positive and negative values. The update addressed the needs of distributed and dynamic-shape workloads by allowing more flexible axis specifications. Regression tests were added to confirm that negative-axis results matched those of positive axes, improving reliability. The feature was developed using Python and JAX, with a focus on numerical computing and parallel programming best practices throughout the implementation process.