Build Pipelines
Financial Document Table Extraction
Table Transformer + TrOCR detecting and extracting structured tables from scanned PDFs with 96%+ accuracy.
The problem
Financial tables in scanned PDFs, with merged cells and nested structures, resist standard OCR and don't come out as usable structured data.
- A merged cell turns one number into three garbled fragments.
- Someone still re-types figures from a PDF into a spreadsheet by hand.
- OCR gets the digits right but loses which row and column they belonged to.
What NEO built
NEO combined Table Transformer for table detection with TrOCR for cell-level OCR, post-processing the result into clean CSV/JSON.
The result
96%+ extraction accuracy
Hits 96%+ table extraction accuracy across multi-page financial documents, including irregular table structures.
From the blog · 8 min
Automated Table Extraction from Financial Documents Using Transformer Models
A deep dive into our production financial OCR pipeline that uses Microsoft Table Transformer and TrOCR to detect, extract, and validate structured tables from PDFs and scanned documents with 96%+ accuracy.
Try this in your workspace
Paste this into NEO chat to kick off the same workflow on your own data.
Extract the tables out of these scanned financial PDFs, even the ones with merged cells, and give me clean structured CSV/JSON instead of raw OCR text.
Paste it in · review the plan · get the diff
Get NEO
