Worked on the mastra-ai/mastra repository to refactor agent model initialization, focusing on improving maintainability and scalability within an agent-based AI system. The approach replaced direct ai-sdk imports with a string-based model name routed through a centralized model router, decoupling model selection from external dependencies. This architectural change, implemented in TypeScript, enhanced code clarity and enabled easier experimentation with different models, including readiness for A/B testing. The update streamlined the initialization path, making future model swaps more straightforward and reducing technical debt. The work demonstrated a strong grasp of AI development principles and modern TypeScript practices in a collaborative environment.