Has anyone struggled with handling custom user code in their LLM application? Came across this use case while trying to enable user-defined LLM output validation functions for my LLM infra project. Would love to talk to anyone that's facing this issue and interested in a solution