Gemini 2.0 Flash PDF Table Extraction

I’m working on a project using 2.0 Flash that’s attempting to extract specific manufacturer and model numbers from tables nested within PDFs and create a structured output that can be used to hit an endpoint in a workflow. Thus far we’re getting really good results however we’re noticing that the output is often missing 1 or 2 Manufacturer/model numbers if there’s multiple referenced in a row. Is this something that is a prompt issue or do we need to fine tune to get better accuracy?

Screenshot for reference showing the row that has multiple manufacturer/model numbers referenced and the return from the query. It misses the Symmons item being requested in the very last cell of the row. (Output below)

LAV-1	American Standard	AQUALLYN 0476.28
LAV-1	Kohler	K-15171
LAV-1	SLOAN	EBF-650