Workflows

Oracle Autonomous AI Vector Database supports two different workflows: integrated embedding and bring-your-own vectors.

Integrated Embedding

The integrated embedding workflow is dependant on the following steps:

  1. Create a vector table that is integrated with an ONNX embedding model loaded into the database.

  2. Upsert a record. Part of the record contains the content that will be automatically be converted to a vector.

  3. Search the vector table for records, specifying how many results should be returned (top K) and any metadata filtering to be used.

Bring-Your-Own Vectors

In this workflow, vectors are generated externally, as described in the following steps:

  1. Create a vector table with a dense vector type.

  2. Generate vectors using your embedding model externally.

  3. Upsert records that include the vector values, plus any metadata on which you would like to filter.

  4. Search the vector table for records, specifying how many results should be returned (top K) and any metadata filtering to be used.

Each workflow can include an optional final step of reranking the results using either an external reranking model or one that has been loaded into the database.

See Create a Vector Table for information about table creation for either workflow type using the create_vector_table() API.