Worked on the AllenDowney/pymc repository to implement a targeted feature optimizing JAX-based sampling workflows. Developed and integrated a logp_fn parameter into the sample_jax_nuts function, ensuring the log probability function is jaxified only once per sampling run. This approach reduced redundant computation and improved overall sampling efficiency, particularly for production workloads. The work involved careful API design, clear documentation of parameter usage, and a focus on performance-oriented refactoring. Leveraged Python, JAX, and numerical methods to streamline probabilistic programming tasks, laying the foundation for faster and more repeatable sampling processes without introducing new bugs during the development period.