Worked on enhancing prompt templates for the InferESG repository, focusing on refining AI agent output to deliver more structured and informative responses. The approach involved improving agent selection logic and restructuring JSON output to better capture user intents, using Python and Jinja for backend development and prompt engineering. New fields for result and query types were introduced, question categories were expanded, and detailed examples for ESG score queries and data manipulation tasks were added. These changes aimed to improve data quality and user experience, demonstrating depth in natural language processing and backend integration, though no bug fixes were reported during this period.