Worked on the APPFL/APPFL repository to establish foundational project scaffolding focused on reproducibility and GPU-accelerated experimentation. Built automated CI/CD pipelines using Docker and GitHub Actions, and set up Sphinx-based documentation to standardize development and testing processes. Developed a GPU-accelerated federated learning simulation by reorganizing the simulation directory, updating experiment configurations, and migrating execution to CUDA with sample-size based client weighting. Addressed robustness and reproducibility in external dataset support, ensuring reliable integration. Leveraged Python, Shell, and YAML to implement infrastructure and documentation improvements, delivering two core features that enhanced the project’s reliability and streamlined collaborative development within one month.