Implement

To implement this solution, you must create your RAG knowledge base and verify your RAG search.

Create a Dify RAG Knowledge Base and Write Data

With Dify Knowledge Pipeline, you can quickly build a RAG application for enterprise private data.
Follow these steps for creating the Dify knowledge base:
  1. To upload a multimodal file, log in to the Dify console, go to the Knowledge Base module, and click Add File to upload a multimodal document (for example, test_cn.pdf containing text and diagrams).
    Dify automatically segments the file into semantic chunks and converts them into vector data.
  2. Verify vector write in Oracle AI Database 26ai.
  3. Vector data is synchronized to Oracle AI Database 26ai. After logging in to the database, execute the following SQL:
    select * from cat;
  4. You should see generated vector index–related tables.
    The following indicates that the data was written successfully: 
    DR$IDX_DOCS_EMBEDDING_VECTOR_INDEX_4B55F797_89B4_4EEF_832C_FF2495C42CCC_NODE$I

Verify RAG Search Results

Verify that Oracle AI Database 26ai ensures search efficiency and accuracy. Oracle AI Database 26ai stores vector data.
Follow these steps to run a recall test in the Dify Knowledge Base:
  1. In the Recall Test function, enter a query keyword such as Oracle 26ai vector features.
  2. The system retrieves relevant vector snippets from Oracle AI Database 26ai and returns results with similarity scores. Example matches:
    • AI vector search built into Oracle AI Database and leveraging partitioning, RAC, sharding, and Exadata for industrial-grade scalability (SCORE 0.65)
    • Oracle AI Database 26ai supports a variety of vector operations (SCORE 0.64)