Georgios Papairo worked on performance optimization for the ggml-org/llama.cpp repository, focusing on GPU programming and SYCL with C++. He developed the Q8_0 reorder optimization specifically for Intel Arc GPUs, extending existing optimization techniques to the Q8_0 data path. This work resulted in approximately threefold throughput improvement and increased bandwidth utilization for large language models such as Qwen3.5-27B. Georgios also addressed a type-check issue in the SYCL backend initialization, ensuring the new optimization activated correctly on real hardware. His contributions expanded Arc GPU acceleration coverage, delivering higher inference throughput and lower latency for supported models in production environments.