-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Unknown
-
None
-
Affects Version/s: None
-
None
The text-to-mql feature in LangChain uses naive sampling for generating the user schema for their collections. This schema is part of the prompt prepared for the LLM that generated the MQL from the user's natural language.
We want to integrate improved schema sampling to better account for the variability in the user's schema, which is typical in MongoDB.
Ideally, this should be available in a library that can be leveraged in various AI frameworks with this functionality. Currently, it is langchain, shortly, we will have CrewAI and MCP.