During April 2025, Wenjie focused on enhancing the reliability and observability of the verl-deepresearch repository by addressing a critical issue in experiment metric tracking. Wenjie identified that MLflow logging was failing when metric names contained special characters, specifically the '@' symbol, which violated MLflow’s naming constraints. To resolve this, Wenjie implemented a Python-based solution that sanitized metric names by replacing '@' with '_at_' before logging, ensuring seamless integration with MLflow and consistent data tracking across experiments. This work emphasized stability and maintainability, leveraging skills in Python and data tracking, and contributed to improved third-party integration compliance for the project.