Developed a feature for the google/langfun repository to enhance token usage visibility and cost accuracy for language model clients. Leveraging Python and backend development skills, introduced cached prompt token tracking within LMSamplingUsage and integrated this data into the pricing logic for both Gemini and general models. Updated cost estimation and pricing information to reflect cached token usage, resolving previous misalignments and improving billing transparency. Focused on robust data modeling and unit testing to ensure accuracy and maintainability, the work enabled more precise budgeting and forecasting for users. Demonstrated end-to-end ownership of usage instrumentation and cross-model pricing integration throughout the project.