Conner focused on low-level kernel optimization for the ThunderKittens repository, where he increased the mla_decode kernel page size from 64 to 256. This change aligned the cache tile layout and reduced the number of iterations and masking required for attention blocks, directly improving performance. He updated all related tests to ensure correctness across affected code paths, demonstrating careful attention to reliability. Working primarily with CUDA and C++, Conner applied his skills in kernel development and performance optimization to deliver a targeted feature that addressed both computational efficiency and maintainability. The work reflected a deep, code-level understanding of system performance.