Description

This PR enhances the SAP HANA Cloud Vector Store integration in langchainjs with three major features that align its behavior with the existing Python implementation:

Enhancements

1. Dedicated Metadata Columns
   Metadata fields can now be stored in dedicated table columns in addition to the JSON metadata field. This enables more efficient filtering and indexing, and mirrors the behavior of the Python implementation (add_texts in hanavector.py).

2. Keyword Search Filtering
   Added support for a $contains operator in metadata filters, which constructs SQL full-text search conditions such as:

   SCORE(? IN ("<column>" EXACT SEARCH MODE 'text')) > 0

   When filtering on fields stored in the JSON metadata, a WITH clause is used to project the relevant key into its own column.

3. Internal Embedding Support
   Introduced a HanaInternalEmbeddings class that uses SAP HANA's native VECTOR_EMBEDDING function to compute document and query embeddings directly in the database. This allows fully in-database similarity search with reduced external dependencies.
   Depending on whether internal or external embeddings are used, the query structure dynamically switches between injecting a precomputed vector (via embedQuery()) or calling the native HANA VECTOR_EMBEDDING function inline.

Tests & Documentation

• Integration tests and documentation have been updated to reflect the new features.
• This implementation brings the langchainjs SAP HANA Vector Store in alignment with the external LangChain Integration Package - Python.

Issue: No associated issue.

Dependencies: No new dependencies were introduced in this PR.

Twitter handle: @sapopensource