Digitise a document
Digitise converts a whole document (a printed page, a handwritten form, a scanned book, a historic manuscript) into clean, structured text. Headlines, section titles, paragraphs, tables, and images are all identified and preserved, so the output is ready for LLMs, RAG pipelines, publishing, or archives.
Digitise is not only for scans. Its typical use cases include:
Reports, product briefs, statements, and multi-page PDFs.
Patient notes, insurance forms, application submissions, personal letters.
Property titles, government registration papers, and legal filings, often in regional scripts.
19th-century manuscripts, Gujarati / Bengali / Tamil colonial-era records, court files, and out-of-print books.
How it works
Every Digitise project follows the same three-step flow.
Upload documents
Drag and drop or click to upload PDF, JPEG, or PNG files, up to 50 MB per file and 10 pages per project.
Step 1: Open Digitise and upload your document
Click Digitise under Products in the sidebar. The Digitise page shows your Digitise configs and recent Digitise projects. Click Upload files, select your document in the file picker, and click Open.
Supported formats: PDF, JPEG, PNG. Limits: 50 MB per file, 10 pages per project. For longer documents, split into 10-page batches before uploading.
Three ready-made Digitise templates ship out of the box under Workspace → Configs → Templates → Digitise:
- Land Deeds Digitisation for property registration papers and legal records.
- Digitise Handwritten Document for scanned handwritten forms, patient notes, and letters.
- Digitise Historic Gujarati Document for regional-language historic manuscripts.
Duplicate any template into your own Config and tweak language or document format as needed.
Step 2: Configure the project
Once the file uploads, the New Project dialog shows a preview of your document on the left and three fields on the right.
Name the project
Project name. A short identifier like Test 1, Court ruling 1987, or Gujarati Land Deed 22-A. Shows up in the Projects list.
Pick the document format
Document format. Choose the option that matches the source: Printed for typeset books, statements, and product reports; Handwritten for handwritten forms, manuscripts, and personal notes; or Mixed (handwritten & printed) for forms with typed labels and handwritten answers.
Step 3: Track processing
The project appears in your list and progresses through Draft → Processing → Completed (or Failed if something went wrong). A “Project submitted, we’ll notify you when it’s ready” toast confirms the run. Single-page documents typically finish in seconds; long PDFs take proportionally longer.
Step 4: Review and edit the output
Click the project to open the editor.
Editor layout
Your document rendered at full resolution with color-coded bounding boxes on every detected element (headers, paragraphs, images, tables, footnotes). Zoom is controlled from the pane’s zoom dropdown (defaults to 100%).
Every detected element as a numbered card, ordered top-to-bottom. Each card has a type dropdown and the extracted content underneath. Tables render inline with the data preserved.
The Sections and Digitised text tabs
The right pane has two tabs:
Sections (default)
Digitised text
Numbered, editable list of every element Sarvam detected. Each section has:
- A type dropdown to reclassify by picking from the list of section types (see below).
- The content. Click to edit inline. Text sections edit as plain text; Table sections render as a proper grid and are editable cell-by-cell.
Section types
Sarvam classifies every detected region into one of the following types. Change the type on any section with the type dropdown, or use it when drawing a new box (see below).
Editing tables with the Table Editor
Double-click any table on the right pane (the rendered grid inside its Table section) to open the dedicated Table Editor modal. It’s the fastest way to review and fix a tabular region without losing the surrounding page context.
Zoom the source preview if needed
Use the Image Zoom dropdown (90%, 75%, and other steps) on the toolbar to verify values against dense statements.
The Table Editor is especially useful for bank statements and invoices where a handful of cells (currency symbols, OCR-misread digits) benefit from a quick human touch-up before export.
Add missing regions with Add boxes
If Sarvam missed a region (a caption, a footnote, an extra column), you can draw it in.
Draw a rectangle
Click and drag on the source page to draw a box around the region you want to capture.
Add boxes is available on Digitise projects only. Extract does not use bounding boxes.
Navigate multi-page documents
For multi-page uploads, use the page selector on the source pane (< 1/N >). Each page’s bounding boxes are shown when that page is active; the Sections pane scrolls to the sections belonging to the current page.
The Sections tab shows a running count in its label (e.g. Sections 77). That’s the total across every page in the document.
View the underlying Config
Click View config in the editor toolbar to see the Config that drove this project: its Document format, Document language, and any advanced settings. Handy when you’re troubleshooting output that doesn’t match expectations.
Step 5: Download
When you’re happy with the output, click Download in the editor toolbar. Choose your format from the dropdown:
Preserves rich layout (headings, images, tables, colored spans). Best for publishing to the web or feeding into any HTML-aware tool.
Clean, LLM-friendly text with heading levels and tables. Best for RAG pipelines, embedding, and archival.
Native Word format. Best when the downstream user will hand-edit the document further.
Flat text without formatting. Best for lightweight indexing and search.
Feeding clean output into an LLM or RAG pipeline
Digitised structured text is a much better LLM input than raw OCR:
- Headings become natural chunk boundaries. Split on section titles for coherent chunks.
- Tables stay parseable. Sarvam preserves them as tables, not as jumbled words.
- Reading order matches expectations. No column-mixing on multi-column layouts.
If your RAG pipeline currently ingests raw OCR, swap in a Digitise Markdown export and re-measure retrieval quality. Most teams see immediate improvement.
Next: Extract as CSV/Excel.