Worked on the Arize-ai/phoenix repository to enhance the robustness of model invocation in backend systems. Addressed a recurring issue where None-valued parameters, specifically top_p and temperature, were being sent to the AnthropicModel API, leading to avoidable errors and experiment interruptions. Applied defensive programming techniques in Python to ensure these parameters are only included when valid, thereby reducing API failures and stabilizing production experiments. This backend development effort improved the reliability of model calls, lowered support overhead, and contributed to smoother onboarding for new contributors, demonstrating a focus on model optimization and resilient API integration within the project.