Elysia Tools
Navigation
Workflow Playbook
Extract OCR text from scans and convert document pages or images into Markdown, JSON, tables, captions, and retrieval-ready chunks with quality checks.
Hubs
A reliable OCR workflow begins with the source type, not with the output you hope to get. Some PDFs are scan-only and need OCR on every page. Others already contain a text layer, or mix digital pages with scanned inserts. That difference decides whether you should run OCR, extract directly, or split the file first.
Narrative reports often need Markdown or clean text so editors, reviewers, or language models can read them. Receipts and ID cards usually need field-level JSON. Financial statements, shipping manifests, and lab reports may depend on table extraction instead of paragraph reconstruction. A single plain-text export rarely serves all of these equally well.
Keep page boundaries, headings, and key values visible while converting the source into richer forms such as tables, captions, or structural JSON. That makes it easier to prove where a total, name, date, or caption came from when someone audits the result later.
OCR is not finished when text appears. Review representative pages and compare headings, totals, names, dates, table headers, and truncated lines against the source. Only after that should you create citation-ready chunks for RAG or publish the extracted Markdown and JSON.
Workflow playbook
Inspect whether the file needs OCR everywhere or only on selected pages, then isolate the relevant range before heavier extraction.
Run OCR for scanned pages or direct text extraction for PDFs that already contain a usable text layer, keeping the raw output for comparison.
Convert the document into richer outputs such as JSON structure, tables, receipts, ID fields, or page-image captions when the downstream task needs explicit semantics.
Package the reviewed content into citation-ready chunks and compare key headings, totals, names, dates, and tables against the source before ingesting it into search or RAG.
Use OCR for image-only pages; prefer direct extraction when the PDF already carries reliable selectable text.
Choose specialized JSON extraction for receipts and ID cards, table extraction for tabular pages, and Markdown or clean text for narrative documents.
Trim the page range before extraction when appendices, covers, or low-value pages would dilute quality and review time.