Worked on enhancing the Claude Haiku model’s reasoning capabilities within the BerriAI/litellm repository, focusing on enabling more robust multi-step task handling and improving automation workflows. The engineering approach centered on strengthening model reasoning, stability, and traceability, with updates implemented using JSON for data management and model optimization. Addressed the integration path for reasoning to ensure stable processing of complex tasks, reducing the need for manual intervention. This work allowed the model to support more advanced automation scenarios across client use cases, reflecting a targeted application of AI development skills to improve both reliability and the overall effectiveness of the system.