Bring catalog-maintained business definitions and object descriptions into natural language queries in Oracle Autonomous AI Database
Business metadata can help ground natural-language SQL in the shared definitions and terminology of your enterprise. A data catalog can hold the business definitions that database metadata alone cannot express: approved meanings for measures, descriptions of data assets, and terms your teams use in reports. Oracle Select AI can use that information as semantic context when translating natural language into SQL. By connecting Informatica Data Catalog as a catalog semantic source, you can bring catalog-maintained context into an AI profile alongside the metadata for the Oracle Autonomous AI Database objects the profile can query.
Use catalog definitions as query context
Consider a catalog definition for an “at-risk customer”: a high-value customer with no order in the past 90 days and an open support escalation. A database column such as SUPPORT_STATUS does not explain that complete business rule on its own. The catalog can provide the approved term and related asset descriptions; database metadata gives Oracle Select AI the names and structure of the tables and columns. Together, this context can help the LLM interpret a request such as “List at-risk customers by region” using the meaning your organization maintains. The generated SQL should be reviewed to confirm that it reflects the intended business definition.
The Informatica integration connects a named catalog as an Oracle Select AI semantic source object. You can then specify that semantic source in your AI profile. At prompt time, Oracle Select AI can retrieve relevant semantic context from configured sources to supplement database metadata. The catalog remains the place where stewards maintain the definitions, while the profile and Select AI determine which catalog context is needed for the query.
Connect Informatica Data Catalog
Create a catalog semantic source with DBMS_AI_SEM.CREATE_SEMANTIC_SOURCE. The example below identifies Informatica as the provider and requests both object-level and global context with the all scope. Replace the endpoint, credential, and data source values with the settings for your environment. The data_source_name should identify the Informatica catalog data source that contains the metadata you want Oracle Select AI to use.
BEGIN
DBMS_AI_SEM.CREATE_SEMANTIC_SOURCE(
source_name => ‘INFORMATICA_SALES_CATALOG’,
type => ‘catalog’,
attributes => ‘{
“provider”: “informatica”,
“scope”: “all”,
“source_endpoint”: “<INFORMATICA_ENDPOINT>”,
“credential_name”: “<INFORMATICA_CREDENTIAL>”,
“data_source_name”: “SALES_DATA_SOURCE”
}’
);
END;
/
The source name is the identifier you reference from an AI profile. The credential name points to the credential configured for access to the catalog endpoint. Manage the credential according to the applicable access-control configuration and grant other profile owners access to the semantic source when they need to use it. A grant enables use of the source; it does not transfer ownership or the credential.
Attach the source to an AI profile
Creating the semantic source does not automatically add it to every profile. Include its name in the semantic_source attribute of the profile used for the request. The following example shows the relevant profile configuration; retain the LLM provider and credential values that apply to your environment.
BEGIN
DBMS_CLOUD_AI.CREATE_PROFILE(
profile_name => ‘SALES_PROFILE’,
attributes => ‘{
“provider”: “<LLM_PROVIDER>”,
“credential_name”: “<LLM_CREDENTIAL>”,
“semantic_source”: [“INFORMATICA_SALES_CATALOG”]
}’
);
END;
/
A profile can reference semantic sources owned by its owner or sources for which the owner has been granted access. If you already manage the profile, add the semantic source through your profile-management workflow. A source cannot be dropped while it is attached to an active profile, so remove the profile reference before retiring the source.
Inspect context and generated SQL
Select AI gives you the ability to inspect the semantic context being sent to the LLM as well as the generated augmented prompt and generated query. As part of your testing, you can inspect the semantic context Oracle Select AI attaches to a representative prompt. Using the Select AI actions SHOWPROMPT, you can display the augmented prompt, including object and column descriptions. Using SHOWCONTEXT, you can display both object and global business semantic context associated with the request.
SELECT AI SHOWPROMPT ‘List at-risk customers by region’
SELECT AI SHOWCONTEXT ‘List at-risk customers by region’
SELECT AI SHOWSQL ‘List at-risk customers by region’
This enables you to check that the expected Informatica definitions appear as expected and that they apply to the intended database objects. Using SHOWSQL, inspect the SQL generated by the LLM, including any filters and joins. Compare the same prompt with and without the catalog source to evaluate the potential effect of your catalog semantic source.
Start with one governed business term
Choose a business phrase that people use often but that has led to inconsistent query results. Confirm that Informatica Data Catalog contains the intended definition and the descriptions needed to relate it to Oracle Autonomous AI Database objects. Configure the catalog semantic source, specify it in the appropriate AI profile, and inspect the context and SQL for that phrase. You can then expand the profile’s catalog coverage with a clearer view of how catalog-maintained metadata can inform natural language queries.
Learn more