Julian worked on the ScrollPrize/villa repository, focusing on improving the inference pipeline by replacing a hardcoded start index with a dynamic, configurable parameter. Using Python and leveraging skills in computer vision and deep learning, Julian introduced a runtime argument, start_idx, managed through InferenceArgumentParser. This change allowed the inference process to adapt flexibly to different deployment environments without manual code edits. By propagating the new argument throughout the codebase, Julian enhanced both maintainability and testability, reducing the risk of misconfiguration. The work addressed a specific bug, reflecting a targeted and thoughtful approach to improving deployment safety and operational flexibility.