Petr Jirsa enhanced AI observability in the microsoft/semantic-kernel-java repository by standardizing telemetry usage metrics for chat completions. He aligned span attribute names and metric conventions with OpenTelemetry best practices, renaming token usage metrics to improve data consistency and analytics accuracy. Using Java and leveraging skills in telemetry and AI services, Petr updated both the instrumentation and related test assertions to ensure reliable reporting and prevent regressions. His work improved the quality of usage analytics and cost modeling, enabling more accurate monitoring of AI service interactions. The changes provided a foundation for consistent, data-driven observability across chat workflows.