During February 2026, Hiso developed a performance-oriented feature for the NVIDIA/NeMo-RL repository, focusing on speculative decoding to enhance post-training workflows. Leveraging deep learning and machine learning expertise in Python, Hiso implemented speculative decoding logic and integrated instrumentation to measure its impact on model performance. The work included updating model configuration files to enable the new feature, ensuring it could be easily adopted in production settings. While no bug fixes were recorded during this period, the contribution demonstrated depth in both algorithmic implementation and evaluation, providing a measurable improvement to post-training processes within the NeMo-RL codebase.