This is illustrated in the following diagram:

Oracle AI Vector Search is designed for Artificial Intelligence (AI) workloads. It allows you to query data based on semantics and image similarity rather than simply keywords. The preceding diagram shows the possible steps you must take to manage vector embeddings with Oracle AI Vector Search.

Primary workflow steps:

1. Generate Vector Embeddings from Your Unstructured Data

   You can perform this step either outside or within Oracle Database. To perform this step inside Oracle Database, you must first import a vector embedding model using the ONNX standard. Your unstructured data can reside within or outside Oracle Database.

   For more information, see Generate Vector Embeddings.

2. Store Vector Embeddings, Unstructured Data, and Relational Business Data in Oracle Database

   After you have generated the vector embeddings, you can store them along with the corresponding unstructured and relational business data. If vector embeddings are stored outside Oracle Database, you can use SQL*Loader or Data Pump to load the vector embedding inside a relational table within Oracle Database. It is also possible to access vector embeddings stored outside the database through external tables.

   For more information, see Store Vector Embeddings.

3. Create Vector Indexes and Hybrid Vector Indexes

   Similar to how you create indexes on regular table columns, you can create vector indexes on vector embeddings, and you can create hybrid vector indexes (a combination of Oracle Text index and vector index) on your unstructured data. This is beneficial for running similarity searches over huge vector spaces.

   For more information, see Create Vector Indexes and Hybrid Vector Indexes.

4. Query Data with Similarity and Hybrid Searches

   You can then use Oracle AI Vector Search native SQL operations to combine similarity with traditional relational key searches. In addition, you can run hybrid searches, an advanced information retrieval technique that combines both the similarity and keyword searches to achieve highly relevant search results. SQL and PL/SQL provide powerful utilities to transform unstructured data, such as documents, into chunks before generating vector embeddings on each chunk.

   For more information, see Query Data With Similarity and Hybrid Searches and Supported Clients and Languages.

5. Generate a Prompt and Send it to an LLM for a Full RAG Inference

   You can use vector utility PL/SQL APIs for prompting large language models (LLMs) with textual prompts and images using LLM-powered interfaces. LLMs inherently lack the ability to access or incorporate new information after their training cutoff. By providing your LLM with up-to-date facts from your company, you can minimize the probability that an LLM will make up answers (hallucinate). Retrieval Augmented Generation (RAG) is an approach developed to address the limitations of LLMs. RAG combines the strengths of pretrained language models, including reranking ones, with the ability to retrieve information from a dataset or database in real time during the generation of responses. Oracle AI Vector Search enables RAG and LLM integration using popular frameworks like LangChain, Ollama, and LlamaIndex.

   For more information, see Work with LLM-Powered APIs and Retrieval Augmented Generation.