AI / ML platforms: fewer indexes, fewer copies, clearer recall policy.

AI / ML

Consolidate the serving layer around embeddings and features.

Vector search

Replace separate ANN indexes with latent scan + SQ8 rerank where the Atlas says the workload recovers.

Feature serving

Serve features from the same representation used for query and cache surfaces.

RAG infrastructure

Reduce duplicated stores around embeddings, metadata, search, and cache.

Proof packs

Use HX-Provenance for signed benchmark and evaluation receipts.