[lordofmoria]

Since LLMs are bad at the null hypothesis (in this case, when a key does not exist in the source JSON), how does this prevent hallucinating transformations for missing keys?

[mpeg]

This isn't using an LLM, it simply checks for similarity between keys using vector embeddings