Fireside Chat: Semantic Mapping

AI frameworks do not always speak the same telemetry language. OpenTelemetry GenAI, OpenInference, Vercel AI SDK, and others can use different attribute names for the same underlying concepts, making consistent observability harder than it should be. Fiddler’s server-side semantic mapping maps those attributes to common concepts, so dashboards, alerts, metrics, and evaluators can work consistently across frameworks, without requiring teams to modify their instrumentation or add client side transformation logic. In our latest fireside chat, we explore why this matters and how semantic mapping can make AI observability more portable across your stack. Watch the conversation to learn more. Learn more: 📖 Blog: Server-Side Semantic Mapping for Consistent AI Observability Across Frameworks: https://www.fiddler.ai/blog/server-side-semantic-mapping?utm_source=youtube&utm_medium=organic_social&utm_campaign=fireside-chats&utm_content=semantic-mappings 📚 Docs: Semantic Mappings: https://docs.fiddler.ai/concepts/semantic-mappings?utm_source=youtube&utm_medium=organic_social&utm_campaign=fireside-chats&utm_content=semantic-mappings 00:00 Introductions 01:32 The Pain of AI Telemetry Fragmentation 02:45 Why AI Observability Is Harder Than Microservices 06:01 What Semantic Mapping Is and How It Works 07:58 Preserving Raw Telemetry (Additive, Not Lossy) 09:28 Operationalizing a Semantic Layer Across Teams 12:20 Where OTel GenAI Conventions Are Headed 14:48 Closing Remarks