PROCESSING COMPLEX DATA STRUCTURES WITH LANGUAGE MODELS

A method including receiving a complex data structure including a combination of text and an arithmetic function. A normalized data structure is generated by normalizing the complex data structure against a static data structure schema. A numerical solver processes the arithmetic function to generate a function output. A prompt is generated by embedding the function output and the normalized data structure into a template prompt. The template prompt includes a first command to a language model to identify a contextual error in the normalized data structure. The template prompt also includes a context that defines a constraint applicable to the complex data structure. The template prompt also includes a second command to generate a model output formatted according to the static data structure schema. The model output includes the contextual error. The model output is generated by processing the prompt with the language model. The model output is returned.