Llama index + SQLAlchemy + Oracle: how to train the data model

Hi there… i was able to use human language to query the database with correct answers:
i.e.1 how many payments are there
i.e.2 how many cross currency payments
i.e.3 how many payments their amount is more than 1 million dollars

But for some functionalities, the table, column and the value doesn’t always use meaningful names
i.e. for entitlements, table name is ENPERUG, column name is PRODCODE, TPCODE

If the user has access to view & change a payment
ENPERUG table will have an entry

COMPANYID, USERID, PRODCODE, TPCODE
00001, 12345, PYMT, VIEW
00001, 12345, PYMT, MODIFY

Is there a way to train the GPT that this table is for entitlement and for these values in the columns means it has access to view and change? I was hoping by uploading an xlsx with tablename, columnname and some description it will derive those info from there to send to LLM for response.

Can someone point me to any articles that talk about how to train the gpt with database model of yours?

Thank you so much.