Developed a token usage analytics feature for image dataset requests in the yhyang201/sglang repository, focusing on providing detailed insights into token consumption for multimodal datasets. Leveraged Python programming and data analysis skills to implement a granular token breakdown, enhancing visibility into resource usage and supporting improved cost and performance analytics. Addressed a dataset token statistics bug related to random multimodal image datasets, ensuring the accuracy of analytics and reliability of downstream reporting. Demonstrated an end-to-end approach by delivering the feature from initial implementation through validation, with particular attention to handling multimodal data tokenization patterns and improving overall observability.