During February 2025, contributed to the instructlab/instructlab repository by developing a configurable performance knob for the SDG pipeline, specifically enabling batch size tuning through system profiles. This work involved updating Python and YAML-based configuration models and the CLI to expose and persist the new batch size parameter, which defaults to 8 for optimized vLLM performance. The approach emphasized configuration management and system administration, allowing operational teams to adjust batch sizes without code changes. By focusing on maintainability and business value, the changes improved runtime efficiency and scalability, supporting faster experimentation and more granular control over resource utilization in large language model workflows.