During December 2025, Li On developed audio transcription token usage tracking for the openai/openai-agents-js repository. By extending the input audio transcription completed event with a usage field, Li enabled detailed analytics and monitoring of token consumption across transcription workloads. This feature, implemented using TypeScript within an event-driven architecture, allows teams to gain cost awareness and supports data-driven capacity planning. Li collaborated closely with Kazuhiro Sera to ensure robust code quality and clear instrumentation. The work focused on full stack development, enhancing observability for audio transcription processes and providing actionable insights for both engineering and operational stakeholders.