I'd like to contribute a blog post and am looking for guidance from SIG Communication and a sponsor from the Semantic Conventions / GenAI SIG if this is a good fit.
Proposed title: Applying the GenAI semantic conventions to a self-hosted LLM gateway
Summary: A vendor-neutral, CPU-only walkthrough of instrumenting a FastAPI gateway in front of a local model (Ollama) with the OpenTelemetry GenAI conventions — operation duration, token usage, finish reasons, and a per-request time-to-first-chunk — and routing them through the Collector into Prometheus/Loki/Tempo. It shows what these signals make visible that infrastructure metrics miss, and is explicit that gen_ai.* is Development status (opted in via OTEL_SEMCONV_STABILITY_OPT_IN).
Outline:
- Why infrastructure metrics miss LLM user experience and model behaviour.
- Current maturity of the OpenTelemetry GenAI semantic conventions and explicit opt-in.
- A reproducible FastAPI + Ollama gateway that emits GenAI telemetry.
- Signals captured: operation duration, time to first chunk, token usage, finish reasons, errors, and trace-correlated logs.
- What four labelled scenarios make visible: normal, expensive, low quality, and errors.
- Practical cautions: metric cardinality, content capture, redaction, and treating the work as a demo rather than production guidance.
Technologies (all open source, CNCF preferred): OpenTelemetry, Prometheus, Loki, Tempo, Grafana, Kubernetes (MicroK8s), Ollama.
Related SIG: Semantic Conventions / GenAI.
Sponsor: Not secured yet; requesting guidance toward the right GenAI/Semconv maintainer or approver.
Why it fits the community: practical, reproducible application of the emerging GenAI conventions with an honest account of their maturity. Not a product pitch.
Draft is ready. If SIG Communication agrees this fits the blog, could you point me to the right SIG sponsor or confirm whether I should bring it to an upcoming GenAI/Semconv meeting?
I'd like to contribute a blog post and am looking for guidance from SIG Communication and a sponsor from the Semantic Conventions / GenAI SIG if this is a good fit.
Proposed title: Applying the GenAI semantic conventions to a self-hosted LLM gateway
Summary: A vendor-neutral, CPU-only walkthrough of instrumenting a FastAPI gateway in front of a local model (Ollama) with the OpenTelemetry GenAI conventions — operation duration, token usage, finish reasons, and a per-request time-to-first-chunk — and routing them through the Collector into Prometheus/Loki/Tempo. It shows what these signals make visible that infrastructure metrics miss, and is explicit that
gen_ai.*is Development status (opted in viaOTEL_SEMCONV_STABILITY_OPT_IN).Outline:
Technologies (all open source, CNCF preferred): OpenTelemetry, Prometheus, Loki, Tempo, Grafana, Kubernetes (MicroK8s), Ollama.
Related SIG: Semantic Conventions / GenAI.
Sponsor: Not secured yet; requesting guidance toward the right GenAI/Semconv maintainer or approver.
Why it fits the community: practical, reproducible application of the emerging GenAI conventions with an honest account of their maturity. Not a product pitch.
Draft is ready. If SIG Communication agrees this fits the blog, could you point me to the right SIG sponsor or confirm whether I should bring it to an upcoming GenAI/Semconv meeting?