Understanding the dataset

Our chatbot works with a Brazilian e-commerce dataset from Olist. It contains real (anonymized) data from 2016-2018 with orders, customers, products, sellers, and reviews.

Not at all! That is the whole point of SCHEMA_INFO. The AI gets the full schema injected into its prompt, so it knows exactly which tables and columns exist. You just need to understand the relationships between tables so you can debug when the AI gets a query wrong.

You understand the dataset structure and how tables relate to each other. Next, we will build the schema context string that gives the AI all this knowledge.