Mintlify focused on improving its assistant by analyzing user feedback and enhancing the underlying data infrastructure. The team identified that search quality was the primary weakness, while the overall response quality was satisfactory. To address this, they rebuilt the feedback pipeline and migrated conversation data into ClickHouse, allowing for better analysis of user interactions. The process involved categorizing feedback from negative interactions, with eight distinct categories established. They examined about 100 negative threads to classify issues, noting that some queries were beyond the assistant's capabilities. For example, questions like requesting a 2FA code were classified under “assistantNeededContext,” highlighting a gap in the assistant’s search functionality across the documentation. The team found no significant changes in feedback trends after a model upgrade in October. However, they expanded the insights available to documentation owners, enabling them to better understand user confusion. UI improvements included allowing users to revisit past conversations, adjusting link behavior to keep users within the docs, and refining mobile interactions for smoother use. The team remains open to feedback, encouraging users to submit feature requests to continue enhancing the assistant.