Popular posts

To build a Retrieval-Augmented Generation (RAG) solution combining 23ai Vector Search and OpenAI in Oracle APEX, follow this process:

Overview of RAG

Retrieval-Augmented Generation enhances large language models (LLMs) by providing access to external knowledge via retrieval methods. The idea is to retrieve relevant information before generating a contextually accurate response.

Steps to Implement RAG Solution

1. Query Reduction: Simplify user queries to enhance search accuracy. For example:

   • Original: "Explain APEX Template Components"
   • Reduced: "APEX Template Components" Utilize advanced models like GPT-4o for better performance in query reduction.

2. Run Vector Search: Use the previously established methods from the earlier post to conduct semantic searches on the relevant blog content.

3. Send Instructions to OpenAI: Construct a prompt to guide OpenAI in generating a response based on the user's question and the retrieved blog post chunks. Ensure to include:

   • Instructions for context
   • The original user question
   • Relevant text chunks from the vector search
   • Blog post IDs for source referencing

   Example Prompt:

   INSTRUCTIONS:
   You are an expert on Oracle technology. Answer the users QUESTION: using the BLOG POSTS: text below. Always Include the number enclosed in ##number## to reference sources used from the included BLOG POSTS:. Prioritize the content of the BLOG POSTS: when determining your answer.
   QUESTION: Can APEX be used to Extend Oracle Fusion Applications?
   BLOG POSTS:
   ##2272## I enjoyed returning to the APEX Alpe Adria conference...
   ...
   

4. Supplement with Links: Ensure the AI includes blog post IDs in its response. This allows for easy reference generation with links to the original content.

5. Display Response: Present the final answer to users, incorporating the relevant sources for transparency and follow-up.

Code Example for RAG Function

Here’s a simplified version of a PL/SQL function that encapsulates the RAG logic:

FUNCTION rag_blog_chunks
 (p_question   IN VARCHAR2,
  p_months_ago IN NUMBER,
  p_max_chunks IN NUMBER,
  p_max_bytes  IN NUMBER,
  p_ai_model   IN VARCHAR2,
  p_ai_temp    IN NUMBER,
  p_qr_model   IN VARCHAR2,
  p_ai_qr_temp IN NUMBER) RETURN VARCHAR2 IS

  CURSOR cr_content (cp_question_vector IN VECTOR) IS
    WITH vs AS (
      SELECT cbp.post_id, cbc.id AS chunk_id, cbc.chunk_text,
             vector_distance(cbc.embedding, cp_question_vector) AS vector_distance
      FROM   cnba_post_chunks cbc, cnba_blog_posts cbp
      WHERE  cbp.post_id = cbc.post_id
      AND    cbp.post_published_date_utc > (SYSDATE - NUMTOYMINTERVAL(p_months_ago, 'MONTH')))
    )
    SELECT vs.post_id, vs.chunk_id, vs.chunk_text, vs.vector_distance
    FROM   vs
    ORDER BY vector_distance
    FETCH EXACT FIRST p_max_chunks ROWS ONLY WITH TARGET ACCURACY 95;

  l_question_vector  VECTOR;
  l_chunks           CLOB;
  l_openai_params    JSON;

BEGIN
  -- Capture Parameter Values and execute the retrieval and generation process
  -- ... Additional code logic ...

  RETURN l_generated_response; -- The final response from OpenAI
END;

Conclusion

This RAG architecture leverages both semantic search and the generative capabilities of AI to provide users with precise, contextually informed responses while ensuring proper sourcing of information.