SAP HANA Cloud Vector — RAG patterns

HANA Cloud ships a native vector column type (REAL_VECTOR) and a SQL function VECTOR_EMBEDDING(text, model) that calls AI Core embeddings inline. That means you can ingest, embed, and query without leaving SQL — but the Python clients (langchain-hana / generative-ai-hub-sdk) are usually nicer.

Schema pattern

CREATE TABLE DOCS (
  ID NVARCHAR(64) PRIMARY KEY,
  TEXT NCLOB,
  METADATA NCLOB,                  -- JSON
  EMBEDDING REAL_VECTOR(1536)      -- dim = your embedding model's dim
);

-- Cosine HNSW index — fastest, smallest, default choice.
CREATE HNSW VECTOR INDEX IDX_DOCS_EMB ON DOCS(EMBEDDING) SIMILARITY FUNCTION COSINE_SIMILARITY;

Before ingestion: data preparation