Description
We're Bizz Vital Clinic LLC, and we have an internal tool that lets our team upload financial documents (invoices, balance sheets, bank statements, reports) and ask questions about them. It's built on Python + FastAPI with OpenAI, LangChain, and ChromaDB, and it's already running. The problem is the PDF parsing. The text extraction isn't reliable, especially with tables and numbers. Rows get dropped, columns get scrambled, and some figures come out wrong. Since the parsed text feeds into our chunking and embeddings, bad parsing means bad answers, and with financial data, we can't have the numbers off. We don't need a rebuild. The rest of the app works fine. We just need someone to look at the parsing part, figure out what's going wrong, and fix it so extraction (including tables and numbers) is accurate. Stack: Python, FastAPI, OpenAI API, LangChain, ChromaDB, Docker What we need: - Review the current PDF parsing code - Find out why extraction is inaccurate - Fix it (open to switching libraries if needed: pdfplumber, PyMuPDF, Camelot, Unstructured, etc.) - Make sure tables and numbers come through correctly - Confirm the fix improves answer accuracy without breaking the existing pipeline It's a focused bug fix, but the codebase is fairly large, so you'll need to be comfortable working in existing code. When applying, please share: - A similar PDF parsing/data extraction problem you've handled before - A quick note on how you'd approach diagnosing this