Question 1

What is a searchable PDF?

Accepted Answer

Your scan stays pixel-identical, but an invisible text layer is added underneath. Ctrl+F, select, copy — everything works like on a normal document.

Question 2

How accurate is Hindi OCR?

Accepted Answer

Modern LSTM models read clean printed Hindi very well. For best results scan at 300 DPI and keep the page straight. Handwriting is not supported.

Question 3

What does "Deskew pages" do?

Accepted Answer

Scanners and phone cameras often tilt the page a few degrees. With the switch on, each page that needs OCR is measured (text-line baselines) and rotated back — up to ±10° — before recognition, which reads crooked text noticeably better. The output page is the straightened scan, and its invisible search layer is drawn from that same straightened image, so search highlights land exactly on the words you see. Pages that already have selectable text are never touched.

Question 4

Why is the first run slower?

Accepted Answer

The OCR engine (~4 MB) and your language model download once from this site — never a third party — and are cached for offline use after that.

Question 5

How well does the invoice fill work?

Accepted Answer

It is label-based text heuristics, not magic: fields are read only from lines that clearly label them ("Invoice No", "Due date", "Total", विक्रेता, कुल योग…) and item rows only from clean numeric tails — anything it did not confidently find stays empty, it never guesses or fabricates values. It works best on clear, straight scans of plainly laid-out invoices; on anything else expect partial or no fill. Always review every field in the editor before downloading. The photo or scan never leaves your device, and there is no AI and no cloud — all recognition and extraction run locally in your browser.

PDF OCR

How it works

Frequently asked