AI w bazie danych Oracle: Vector Search, Select AI i Machine Learning (23ai/26ai)

Description and use

Oracle 23ai/26ai lets you store vector data as native columns, perform semantic search with VECTOR_DISTANCE, generate embeddings inside the database, and translate natural language queries to SQL via Select AI.

Problem

Integrating semantic search and machine learning directly inside Oracle Database without exporting data to external tools.

Outcome

Oracle 23ai/26ai enables end‑to‑end AI workflows inside the database, reducing data movement, improving latency, and simplifying architecture.

Workflow

• Define VECTOR column with dimensions and format, e.g., VECTOR(384,FLOAT32).
• Insert or generate embeddings using VECTOR_EMBEDDING on an imported ONNX model.
• Create vector index (HNSW or IVF) on the VECTOR column.
• Query semantic similarity with VECTOR_DISTANCE and optional FETCH APPROX for performance.
• Use Select AI (DBMS_CLOUD_AI) to convert natural language to SQL queries.
• Leverage Oracle Machine Learning for additional ML tasks within SQL/PL-SQL.
• Monitor performance and adjust index parameters as needed.

Evidence

Vector type is a first‑class citizen in Oracle 23ai/26ai, allowing embeddings to live alongside business data.

Source