> Our prompt is currently very simple: "OCR this PDF into this format as specified by this json schema" and didn't require some fancy "prompt engineering" to contort out a result.