Issue: Retrieval Convention for RAG in OpenTelemetry Spec #3021
Description
Issue: Retrieval Convention for RAG in OpenTelemetry Spec
What are you trying to achieve?
I’m trying to understand if there has been any prior work or discussion around defining a standard convention for the retrieval component of RAG (Retrieval-Augmented Generation) systems within OpenTelemetry — similar to existing conventions for other spans (e.g., HTTP, DB, messaging, etc.).
The goal would be to have a consistent way to represent retrieval operations, including:
- Query formulation and embedding generation
- Vector database or index lookup
- Scoring and ranking
- Context selection or filtering
This could enable better observability across RAG pipelines, especially when integrating with LLM inference spans.
What did you expect to see?
I expected to find an existing convention or proposal (e.g., semantic conventions, trace attributes, span names, etc.) covering retrieval or vector store interactions, but couldn’t locate one in the spec repository or related issues.
Additional context
If there’s any ongoing discussion, draft proposal, or related SIG activity, it would be helpful to know where to follow or contribute.
If this is a gap, I’d be happy to help draft a proposal for RAG retrieval conventions.
Tip: React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.