Enhanced Support for Function-Based Queries in Vanna AI

[sinjup]

We are encountering issues when adding function calls and PostgreSQL functions as training data in Vanna AI. Specifically, the model struggles with accurate identification and execution of functions with their parameters in queries. This impacts the accuracy and consistency of responses when function-based queries are involved. We seek guidance and potential enhancements to improve Vanna AI's handling of function-based queries.

Feature Details

1. Function Identification
   Current Issue: Vanna AI struggles to identify functions and understand their parameters in SQL queries.
   Request: Implement enhanced parsing capabilities for recognizing functions, including user-defined functions and built-in PostgreSQL functions, and correctly associating parameters.
2. Function Execution
   Current Issue: Vanna AI does not consistently execute identified functions with the provided parameters.
   Request: Enable the model to support function execution in response to user queries, ensuring parameters are correctly passed and processed.
3. Training Data for Functions
   Current Issue: Inconsistent results when function calls are included in training data.
   Request: Provide best practices and enhancements to the training pipeline to improve Vanna AI’s ability to learn from function-based training data.
4. Performance Impact
   Concern: Will training with functions impact the model’s performance, leading to slower query responses?
   Request: Assess and optimize the model's performance when processing function-based queries and recommend potential workarounds if performance degradation is identified.
   Expected Outcome
   Improved accuracy in identifying and executing functions with parameters.
   Enhanced model consistency when handling function-based queries.
   Clear guidance on how to structure function-based training data for optimal results.
   Impact
   Enhancing Vanna AI's capability to handle functions will significantly improve its utility in technical and database-heavy environments, particularly in scenarios involving complex SQL queries with functions.