Worked on the JuliaGPU/CUDA.jl repository to enhance the correctness and robustness of batched matrix-vector operations, specifically addressing issues in the batched GEMV implementation. Focused on resolving a bug related to transposed matrices and batching, the work involved adding comprehensive tests to validate handling of transposed matrices and various batching scenarios. Ensured that input dimensions remained consistent across batched operations to prevent dimensionality errors, thereby improving reliability for users. Utilized Julia and CUDA, applying expertise in GPU computing and linear algebra to deliver a targeted fix that reduces edge-case failures and strengthens the overall stability of batched GEMV functionality.