Worked on extending the oneapi-src/oneDNN repository to add BF16 data type support for the ACL inner product operation on aarch64 platforms. The approach involved implementing compatibility checks and broadening fast-math conditions to include BF16, targeting improved performance for deep learning workloads on 64-bit ARM architectures. Leveraged expertise in ARM architecture, CPU optimization, and embedded systems, using C++ to ensure the new path integrated cleanly with existing code. All changes were reviewed for maintainability and future extensibility, aligning with repository standards. This work enables a performance-oriented execution path for select models, addressing the need for efficient BF16 computation.