Story

I’m curious how teams are handling answer reuse in production RAG systems.

We’ve looked at: • naive semantic caching • query similarity thresholds • embedding-based reuse

…but correctness risk seems to make most of these approaches scary in practice, especially when: • source docs change • retrieval context shifts • similar queries require different answers

Are teams: • avoiding answer reuse entirely? • limiting reuse to very narrow FAQ-style flows? • using some form of conservative gating or shadow evaluation?

Not looking for vendor recommendations — just trying to understand what’s actually working (or failing) in real systems.

Thanks!