pdf-pipeline
Python · pdfplumber · Pandas · PostgreSQL
Fund disclosure PDFs are where data goes to hide. 489 of them, each formatted by someone with strong and conflicting opinions about tables.
The pipeline
pdfplumber does the first pass; an OCR fallback catches the scanned ones; an LLM-based correction step fixes the extraction errors that slip through. The output lands in PostgreSQL as clean, validated portfolio data.
The payoff
Monthly revenue-sharing reconciliation and algorithmic fund selection now run as code. The previous implementation was a human, an Excel file, and a lost afternoon — every single month.