Worked on the vllm-project/tpu-inference repository to enhance observability for the KV connector by developing a targeted instrumentation feature. This addition exposed queue length metrics, allowing teams to monitor the number of requests being pulled versus those waiting, which supports faster diagnosis and more informed capacity planning. The implementation integrated seamlessly with the existing telemetry framework and adhered to project contribution standards, including signed-off commits. Focused on backend development and metrics monitoring, the work leveraged Python and unit testing to ensure reliability. No major bugs were addressed during this period, as efforts centered on improving monitoring and alerting readiness.