Pen Pals
Signatures of all things I am here to read, seaspawn and seawrack, the nearing tide, that rusty boot.
— James Joyce, Ulysses (1922)
Draw sixteen letters in the air with your phone. This will take you about a minute and a half. Your phone’s sensors record how your hand accelerates and twists, and we turn that motion into centimetres, seconds and degrees — the same signals our neural net reads to turn air-writing into text. Then we compare your writing with the recorded handwriting of famous historical figures and find the hand most like yours.
How to hold itHold your phone roughly flat, like a TV remote, and press the button with your thumb while you draw.
You’re in the LinkedIn app’s built-in browser. It can block the motion sensors the reading needs.
Anonymous — motion only, kept for research.
What we collect
Your phone’s motion and timing while you draw, plus basic device info (screen, platform, timezone). No name, no login, no location, no account. It’s used to study what handwriting motion reveals.
Enable motion
The reading works from how your hand moves, so it needs your phone’s motion sensor.
Open this on your phone
Pen Pals reads how your hand moves through the air, using the motion sensors in a phone. A laptop has none, and a mouse cannot be waved.
Point your phone’s camera at the code, or type the-wand.ai/pen-pal. About a minute and a half.
Now the joined-up half
That was the letters. The rest are whole words, written in one go without lifting — the way you’d write them on paper.
nine words · about ninety seconds
Now on paper
With a black or blue pen, or a pencil, write this on a plain white sheet — in your normal handwriting, at your normal size, near the middle of the sheet:
dear friend, yours truly
Then photograph the whole sheet from directly above, in good light.
Your photo is used to measure the slant, size and shape of your handwriting. It is never published, and it is kept in our private research archive.
Which looks more like your handwriting?
Tap one. There are no right answers — go with your first impression.
Pick the three most like yours
0 of 3 chosen
Reading your hand…
measuring your motion
Your side is drawn from your measurements — size, slant, steadiness, evenness — not from the path your hand took. We can’t recover that from motion alone, and we don’t pretend to. It’s a portrait of your style, not a photograph of your strokes.
How this works
You drew sixteen letters; your phone clocked the motion 60 times a second. Under the hood, the phone’s motion processor fuses its accelerometer and gyroscope — tracking rotation as quaternions — to separate gravity from your hand’s own acceleration. We resolve that into a writing frame, integrate acceleration into velocity, and turn it into physical measurements — centimetres, seconds, degrees — set against 349 people who drew the same letters in our study.
Each figure carries a handwriting profile scored from documented period descriptions and their surviving ink, every quote cited to its source. For tempo, slant and flourish we went further: we traced each figure’s actual signature and re-animated the pen with a motor law measured on our own 52,191 letters — on a 40-writer signature dataset with real timing, the simulation predicts true writing tempo at rank correlation 0.80. Measures that failed that test are kept out of the match.
Your % match is an affinity score, scaled against those 349 real hands: 100 would mean indistinguishable from the figure’s documented profile, and the score halves with each step of typical-person distance. Across our study participants the best match usually lands in the seventies — nobody scores 100 without writing exactly like their figure. When the field is close, the card says so: a hand varies a little from one run to the next, and on a retake a close second can take the top spot.
Two of the seven measures are analogies, named plainly: vigour stands in for pen pressure, and slant in the air for slant on the page.
The wand itself — the device this engine comes from — reads air-writing with a convolutional neural network trained on this corpus. Pen Pals deliberately keeps its matching simpler and fully inspectable: percentiles and weighted distances you could audit by hand.
Some things a hand doesn’t give away — sex is a coin flip in our data, so we print “unreadable” rather than invent a number; the full study of 52,191 air-written letters, including what failed, is on our blog.
This is how the wand works — it learns your hand.
Meet the wand