Late Edition: Giving OCR Text a Second Look

Description and use

Late Edition uses a Gemma model to correct OCR errors, applying strict safety checks to avoid unwanted changes.

Problem

OCR can misread characters, distort words, and turn a perfectly readable page into a messy block of text.

Outcome

32 passing tests and 6 failures; validation thresholds tested on a small sample; model only receives OCR text, not original image.

Workflow

• Split the OCR text into manageable chunks
• Request corrections from a Gemma model via Gemini API
• Validate each proposed correction with deterministic checks (digits, capitalized words, similarity threshold)
• Generate an HTML report showing original text, accepted edits, and rejected suggestions

Evidence

OCR can misread characters, distort words, and turn a perfectly readable page into a messy block of text.

Source

Late Edition takes OCR text, splits it into chunks, and sends those chunks to a selected Gemma model.

Source

Before accepting a proposed correction, it runs a set of deterministic checks to decide whether the change is safe enough to keep.

Source

The validation layer checks three things: Digits, Capitalized words, Text similarity.

Source

My latest test run reported 32 passing tests and 6 failures.

Source

The model only receives OCR text; it never sees the original page image to verify a suspected error.

Source