Worked on improving the reliability of model registration in the mlflow/mlflow repository by addressing a regression related to environment packaging. Focused on backend development using Python, the solution involved implementing a pre-registration cleanup step that removes any existing _databricks directory before a new model registration. This approach ensured idempotent and conflict-free registrations, reducing errors when models are registered multiple times with env_pack. The fix enhanced the stability of the MLflow model registry, supporting smoother CI/CD pipelines and more dependable production deployments. Emphasized thorough testing to validate the changes, resulting in fewer registration-time failures and more robust automated workflows.