Jev: New Guinea

Never going to be able to replace maybe 40-70% of LLM calls for a given pipeline depending on the business task, cutting the API costs on those calls by an order of magnitude.