Worked on enhancing tracing fidelity in the envoyproxy/ai-gateway repository, focusing on backend development and observability for large language model (LLM) contexts. Addressed a key issue by removing the OpenTelemetry Go SDK span attribute limit, allowing complete capture of high-cardinality attributes during long AI conversations. Introduced opt-out controls for sensitive data using environment variables, supporting privacy and operational requirements. Leveraged Go and OpenTelemetry to ensure that production traces are no longer truncated, which reduces gaps in trace data and enables more reliable debugging. This work improved the accuracy and completeness of tracing for multi-turn AI gateway interactions in production environments.