Anton Mitkov enhanced SYCL tensor operations in the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories by optimizing matrix multiplication paths and improving debugging workflows. He removed unnecessary f16 to f32 conversions in DNNL-backed matmul operations, reducing memory overhead and streamlining computation. Anton refactored debug utilities to return strings instead of void, enabling more flexible and detailed logging for copy operations. Working primarily in C++ with a focus on SYCL programming and performance optimization, his contributions improved runtime efficiency and observability. The work demonstrated a strong grasp of backend development and high-performance computing, addressing both performance and maintainability concerns.