Gubbi developed a feature for the google/langfun repository that enhanced token usage visibility and improved cost accuracy for language model services. By introducing cached prompt token tracking within LMSamplingUsage, Gubbi extended the pricing logic to account for cached tokens across both Gemini and general models. This work involved updating data models and integrating new pricing information, ensuring that cost estimation accurately reflected actual usage. Using Python and backend development skills, Gubbi delivered end-to-end instrumentation and pricing integration, which resolved previous misalignments and enabled more precise budgeting. The work demonstrated depth in API development, data modeling, and unit testing practices.