Reddit user seeks a model for bank-statement document intelligence
A Reddit user proposed parsing bank-statement PDFs with OCR, converting the output into Markdown files, and feeding those files to a text model for expense categorization. The post seeks a recommendation among Qwen, Gemma, LFM, or another model but provides no benchmarks, implementation results, or final selection.
The discussion highlights a practical document-intelligence pattern that separates OCR extraction from language-model analysis, while offering no evidence for choosing one model over another.
I'd like to make a pipeline capable of understanding pdfs of bank statements and then group expenses by categories etc. My thought was to first parse them with ocr into .md files and then feed that to a text model. I think it's more lightweight and accurate this way. But then I'm not sure what model to use to feed all these md files to get something good. When accuracy matters, should I go for qwen, gemma, lfm, or something else? Thank you for your help!