Worked on enhancing AI observability in the microsoft/semantic-kernel-java repository by standardizing telemetry usage metrics and improving test reliability. Focused on aligning OpenTelemetry span attribute names for chat completions with industry best practices, renaming metrics to clarify input and output token usage. Implemented these changes in Java, ensuring consistent instrumentation across chat workflows for more accurate analytics and cost modeling. Subsequently updated test assertions to reflect the new telemetry schema, maintaining data integrity and preventing regressions. Leveraged skills in AI services, telemetry, and testing to support reliable monitoring and facilitate cross-team analysis of AI service interactions within the repository.