We generated business context from the open-source ride-hailing app Namma Yatri, and add that context in Snowflake comments, and then measuring the performance of Snowflake Cortex with and without context generated by Zingle. We have selected these 4 questions randomly.

Example 1: Ride Cancellations

Question: What percentage of ride cancellations are initiated by drivers versus customers?

• ❌ Without Zingle: Cortex picks the wrong column and misinterprets initiator role.

• ✅ With Zingle: Picks the correct column. It also explains cases not included in this analysis like system initiated cancellations and unaccepted driver requests.

• Reasoning: Zingle adds process-level context, ensuring correct mapping of columns and informing user exactly about what data answer contain and what it doesn’t contain.

• Answer Without Zingle ❌

• Answer With Zingle ✅

Example 2: Safety Incidents

Question: Which city or routes have most safety incidents?

• ❌ Without Zingle: Cortex tries to generate SQL but fails because the data doesn’t exist in schema.
• ✅ With Zingle: Responds transparently: “This data is not available in your schema.”
• Reasoning: Zingle prevents hallucination as AI agents know about data in depth