Handwriting

Turn handwriting into text.

Know exactly which lines need human review. Line-by-line confidence scores for cursive, print, and mixed scripts.

Try the demoSecurity & privacy
Australian-ownedZero retentionPDF, JPG, PNG, HEIC
89%

Original

Original handwritten math sample

Recognized text

**Trigonometric Identities:** sin⁡300°=sin⁡(3⋅90°+30°)=cos⁡30°=−32\sin 300° = \sin(3 \cdot 90° + 30°) = \cos 30° = -\frac{\sqrt{3}}{2} tan⁡x±tan⁡y=sin⁡(x±y)cos⁡x⋅cos⁡y\tan x \pm \tan y = \frac{\sin(x \pm y)}{\cos x \cdot \cos y} **Pythagorean Theorem:** a2+b2=c2a^2 + b^2 = c^2 **Circle Sector Formulas:** Pn=α360°⋅π⋅r2P_n = \frac{\alpha}{360°} \cdot \pi \cdot r^2 L=α360°⋅2πrL = \frac{\alpha}{360°} \cdot 2\pi r
handwritten-math.jpgTrig equations · LaTeX output

Technology

Built for documents others can't read

Cursive forms, mixed print, faded ink, and low-light photos — processed reliably.

Confidence scores on every line

Know exactly which lines need human review. Every line scored for certainty — no black-box guessing.

Zero document retention

Processed in memory, never stored. No copies, no backups, no training data.

Cursive, print, and mixed scripts

Faded ink, low-light photos, mixed styles on one page. Handles the handwriting others can’t read.

FAQ

Frequently asked questions

01

How does the OCR handle cursive vs print?

The model identifies stroke connectivity patterns to switch decoding strategies mid-line. Cursive ligatures and isolated print characters coexist freely — no segmentation step, no mode toggle. Mixed-style pages decode in a single pass.

02

What is a confidence score?

Every recognised line carries a 0–1 confidence score derived from the decoder’s beam-search probabilities. Low-scoring lines surface automatically, separating high-certainty output from regions that warrant verification. No result is opaque.

03

How does it handle low-quality scans?

An adaptive pre-processing pipeline normalises contrast, corrects skew, and separates ink from background noise. The recognition model was trained on synthetically degraded inputs — faded ink, bleed-through, creased paper, low-light phone captures — so it degrades gracefully rather than failing on imperfect sources.

04

Can it recognise mathematical notation?

Handwritten equations are parsed into structured LaTeX using a spatial-relationship model that resolves fractions, superscripts, subscripts, Greek letters, and operator precedence from 2D layout. The math sample in the demo above was produced by the same pipeline.

How it's built

Built for sensitive documents

Australian infrastructure. Zero retention. Privacy by design.

Australian-owned

Ryder AI Pty Ltd, QLD. Cloudflare edge is global.

Encrypted in transit

TLS 1.2+ via Cloudflare. No document retention.

Line-level confidence

Every line scored 0–1. Low-certainty regions surface for verification.

Zero retention

No storage, no backups, no training data. Privacy by design.

Explore the technology