Vector Utilities
Vector Utility API
The Vector Utility API provides a SQL function VECTOR_CHUNKS which processes text into pieces (chunks) in preparation for the generation of embeddings to be used with a vector index. The API is configurable in terms of size of chunks and rules for splitting chunks.
While it is possible for you to create your own chunking algorithms, utilizing this functionality could save you time and aid in faster development with a pre-packaged SQL function.
Chainable Utility Functions for Vectors
DBMS_VECTOR provides a set of utility functions for processing text for the creation of vector indexes. These functions may be chained together such that the output from one function is used as the input for the next.
This feature offers a straightforward yet very customizable method for you to turn textual content, like a PDF document or VARCHAR2 database field, into the embeddings necessary for a vector index. This capability enables you to seamlessly develop with vectors, facilitating the creation of the next generation of Artificial Intelligence applications with ease.
Vector Format Output for Feature Extraction Algorithm
Feature extraction algorithms produce a set of features that represent projections in a lower dimensional latent space. The output is typically numerical and dense. VECTOR type representation is the natural choice.
Feature extraction algorithms represent a principled approach to vectorizing relational data. The vectorized representation can be used for similarity search.
In-Database Algorithms Support for VECTOR Data Type Predictors
This feature enables users to include one or more columns of VECTOR data type as predictors along with structured enterprise data to in-database machine learning algorithms.
Vector representations of unstructured data can be a powerful input to traditional machine learning algorithms. They enable efficient data processing on text and image data, helping to speed data-driven decision making. Providing vectors as input to machine learning models enables handling a broader class of use cases.
JSON_TEXTCONTAINS and JSON_EXISTS Support in DBMS_HYBRID_VECTOR.SEARCH
The DBMS_HYBRID_VECTOR.SEARCH API now supports a way for users to specify optional JSON_EXISTS and JSON_TEXTCONTAINS clauses as part of their hybrid search.
Writing complex queries requires the ability to further constrain hybrid (semantic and textual) search by JSON_EXISTS and JSON_TEXTCONTAINS when working with complex JSON data. This feature lets you design sophisticated applications that better satisfy users' search criteria.
Select AI with RAG and Synthetic Data Generation
Select AI now supports retrieval augmented generation (RAG), synthetic data generation (SDG), and the AI providers Google, Anthropic, and Hugging Face. These enhancements build on the existing natural language-to-SQL (NL2SQL) capability.
Select AI on Oracle Autonomous AI Database already supports RAG, SDG, and a broader set of AI providers. This feature provides greater feature parity for non-Oracle Autonomous AI Database users. Select AI enables you to easily develop AI-based applications using large language models (LLMs) and transformers from SQL and PL/SQL.
RAG enables you to provide enterprise-specific and more current data to enhance LLM responses and reduce hallucinations, which supports a wider set of use cases.
SDG enables you to generate data. This synthetic data has many use cases. For example, it can help you to jump-start a new project that has no user data, validate the user experience using the application interface, or populate a metadata clone for development and test scenarios.
The additional AI providers give you more options for using LLMs and transformers with Select AI.