During January 2026, Petar Milojevic developed a hardware-aware optimization feature for the tt-inference-server repository, focusing on the Galaxy6U device. He introduced a new device model specification that enabled data parallelism with DP=4 and fine-tuned memory cache settings to improve model inference throughput and latency. His work included integrating the DP4 configuration into the continuous integration pipeline and aligning deployment paths and tests for Galaxy devices. Using Python and leveraging machine learning and model development expertise, Petar’s contribution addressed deployment readiness and performance optimization, demonstrating depth in both system integration and targeted hardware adaptation within a short project timeframe.