XML Formatted Prompts, different providers very different results?
Does Vapi do any specific XML parsing for specific providers? For example an XML formatted prompt works really well for OpenAI models however has issue when using Google models?
Vapi helps developers build, test, and deploy voice agents at scale. We enable everything in between the raw models and production, including telephony, test suites, and real-time analytics.
Vapi doesn't perform specific XML parsing based on the provider. Any differences in results between OpenAI and Google models could be due to how each model interprets the XML content and responds to it. Each model might have unique strengths and limitations in handling structured data like XML. If you're seeing discrepancies, consider adjusting the prompt format for each provider or experimenting with simpler data structures to find a suitable configuration. You can also refer to each provider's documentation for any specifics on data parsing. For more details on configuring providers, you can check the Provider Keys documentation.