Worked on the rapidsai/cugraph repository to enhance the ForceAtlas2 graph layout algorithm by addressing stability issues in the prevent_overlapping mode. Implemented a diminishing speed factor, following the methodology outlined in the ForceAtlas2 paper and Gephi’s reference implementation, to improve convergence and overlap prevention in crowded graph visualizations. This required updates to CUDA kernel functions and adjustments to test parameters, ensuring more reliable and visually stable layouts for analytics workflows. Utilized C++, CUDA, and Python to deliver these improvements, focusing on the quality and reliability of graph algorithms. The work resulted in a targeted bug fix rather than new feature development.