RAG with Vector Index in 26ai

 Updating my previous demo that was in 23ai  to run in  Oracle AI Database 26ai with two enhancements :

vector_memory_size  set to 512MB   (yes, this is a very small on-premises Free 26ai image)

INMEMORY NEIGHBOR GRAPH Index using Hierarchical Navigable Small World (HNSW)

[oracle@localhost ~]$ sqlplus vector_demo/vector_demo

SQL*Plus: Release 23.26.0.0.0 - Production on Sun Nov 16 09:37:39 2025
Version 23.26.0.0.0

Copyright (c) 1982, 2025, Oracle.  All rights reserved.

Last Successful login time: Sun Nov 16 2025 09:32:43 +00:00

Connected to:
Oracle AI Database 26ai Free Release 23.26.0.0.0 - Develop, Learn, and Run for Free
Version 23.26.0.0.0

SQL> set echo on
SQL> !ls *sql
Create_Vector_Index.sql  Query_Vectors.sql

SQL> @Create_Vector_Index.sql
SQL> CREATE VECTOR INDEX my_data_vectors_ndx ON my_data_vectors (sentence_vector)
  2    ORGANIZATION INMEMORY NEIGHBOR GRAPH
  3    DISTANCE COSINE
  4    WITH TARGET ACCURACY 95
  5  /

Index created.

SQL> show parameter vector_memory

NAME                                 TYPE        VALUE
------------------------------------ ----------- ------------------------------
vector_memory_size                   big integer 512M
SQL> @Query_Vectors.sql
SQL> set pages600
SQL> set linesize 156
SQL> col my_sentence format a148 wrap
SQL>
SQL> ACCEPT text_input CHAR PROMPT 'Enter your query : '
Enter your query : image processing
SQL> VARIABLE text_variable VARCHAR2(1000)
SQL> VARIABLE query_vector VECTOR
SQL> BEGIN
  2    :text_variable := '&text_input';
  3    SELECT vector_embedding(ALL_MINILM_L12_V2_AUGMENTED USING  :text_variable as data) into :query_vector;
  4  END;
  5  /
old   2:   :text_variable := '&text_input';
new   2:   :text_variable := 'image processing';

PL/SQL procedure successfully completed.

SQL>
SQL>
SQL> SELECT my_sentence, vector_distance(sentence_vector , :query_vector, COSINE) as Calc_Vector_Distance
  2  FROM my_data_vectors
  3  ORDER BY 2
  4  FETCH FIRST 3 ROWS ONLY;

MY_SENTENCE
----------------------------------------------------------------------------------------------------------------------------------------------------
CALC_VECTOR_DISTANCE
--------------------
      VanceAI.com   image enhancement
           5.16E-001

      Stable Diffusion: An open source model that generates high quality images from text or other images, offering customization and control
           5.51E-001

      Hotpot.ai   AI image editing
          6.109E-001


SQL>
SQL> SELECT my_sentence, vector_distance(sentence_vector , :query_vector, COSINE) as Calc_Vector_Distance
  2  FROM my_data_vectors
  3  ORDER BY 2
  4  FETCH APPROX FIRST 3 ROWS ONLY;

MY_SENTENCE
----------------------------------------------------------------------------------------------------------------------------------------------------
CALC_VECTOR_DISTANCE
--------------------
      VanceAI.com   image enhancement
           5.16E-001

      Stable Diffusion: An open source model that generates high quality images from text or other images, offering customization and control
           5.51E-001

      Hotpot.ai   AI image editing
          6.109E-001


SQL>

SQL> select * from dbms_xplan.display_cursor();

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  1z2ujsrc9xsb0, child number 0
-------------------------------------
SELECT my_sentence, vector_distance(sentence_vector , :query_vector,
COSINE) as Calc_Vector_Distance FROM my_data_vectors ORDER BY 2 FETCH
APPROX FIRST 3 ROWS ONLY

Plan hash value: 3894957757

------------------------------------------------------------------------------------------------------
| Id  | Operation                      | Name                | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT               |                     |       |       |     2 (100)|          |
|*  1 |  COUNT STOPKEY                 |                     |       |       |            |          |
|   2 |   VIEW                         |                     |     3 |  6024 |     2  (50)| 00:00:01 |
|*  3 |    SORT ORDER BY STOPKEY       |                     |     3 |  4938 |     2  (50)| 00:00:01 |
|   4 |     TABLE ACCESS BY INDEX ROWID| MY_DATA_VECTORS     |     3 |  4938 |     1   (0)| 00:00:01 |
|   5 |      VECTOR INDEX HNSW SCAN    | MY_DATA_VECTORS_NDX |     3 |  4938 |     1   (0)| 00:00:01 |
------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter(ROWNUM<=3)
   3 - filter(ROWNUM<=3)


25 rows selected.

SQL>

Here I demonstrate querying  the same set of 130 sentences about AI as in the previous demo, but now with a Vector Index configured as an In-Memory Neighbour Vector Graph Index and a Target Accuracy of 95% based on COSINE Distance.

My next run would be with a much larger data set (instead of just 130 sentences)