Try out serializing with TOON instead of JSON and benchmark speed/accuracy with various models/services

[azimux]

Not sure if the LLM can handle interpreting TOON data with a json schema or not... but worth a try!