Cookie preferences

HundrED uses cookies to enhance user experiences, to personalise content, and analyse our web traffic. By clicking "Accept all" you agree to the use of all cookies, including marketing cookies that may help us deliver personalised marketing content to users. By selecting "Accept necessary" only essential cookies, such as those needed for basic functionality and internal analytics, will be enabled.
For more details, please review our Cookie Policy.
Accept all
Accept necessary
search
clear

Grafari

Adaptive spelling that turns each child's own mistakes into their own practice plan

Spelling practice is usually identical for every child in the room, feedback arrives days later when the exercise book comes back, and the children who struggle get more of the same list. Constructor Grafari is an adaptive spelling and writing application for children aged 8+ that selects every exercise from the individual learner's own error behavior, so practice targets the patterns that child a

Overview

Information on this page is provided by the innovator and has not been evaluated by HundrED.

Updated August 2026
Web presence

2021

Established

4

Countries
Students early
Target group
That no child spends years practicing the same spelling list as everyone else and concluding they are stupid. Written language is the gate to most of the rest of school, and the children who cannot get through it usually get more of the drill that already failed. We would like practice to start from what a particular child gets wrong, and correction to arrive inside the minute rather than inside the week – and we would like the assumption that a struggling speller is careless to be replaced by the understanding that mapping spoken language onto written form is genuinely hard for some brains and needs a different route in. We would also like the idea behind Grafari to become unremarkable: that a word can be presented as color, shape, structure, and sound rather than as black letters on white paper, and that when the conventional route fails a child, the answer is a different route rather than more repetitions of the same one.

About the innovation

Why did you create this innovation?

The work began with one child. In 2002 Markus Gross, professor of computer science at ETH Zurich, began looking for a better way to help his own son, who has dyslexia. That question became a research program, then the ETH spin-off Dybuster AG in 2007, whose first product is now Grafari.

Spelling and writing difficulties are among the most common barriers in primary education, and conventional practice handles them badly. Word lists are identical for every child regardless of which patterns that child actually fails on. Feedback arrives days later when the exercise book comes back, by which point the wrong spelling has been rehearsed. And children who struggle get more of the drill that is already not working, which erodes confidence as fast as it builds skill.

Teachers cannot diagnose per-child error patterns across a whole class by hand, and parents who want to help have no visibility into what is going wrong. For a child with dyslexia the underlying problem is not effort at all, but the difficulty of mapping spoken language onto written form.

The founding insight was that a word does not have to be presented only as black letters on white paper. If each word is encoded as a computed combination of color, shape, structure, and sound, the brain can recruit regions less used in conventional reading and writing – and a child who has failed at the conventional route gets a different one. We built Grafari to make that route available to any child, in 15 minutes a day.

What does your innovation look like in practice?

A child works in short sessions – 15 to 20 minutes, around four times a week, the pattern used in the published studies.

The words in each exercise are not from a class list. They are selected from this child's own error behavior, so a child who reliably fails a pattern meets it until it holds, and does not spend time on words already secure. Each word is presented as a computed combination of color, shape, structure, and sound rather than as plain black letters: letter-and-color matching links sounds to spelling patterns, and words are broken into syllables so they are processed in smaller units.

Correction happens inside the session, not days later. Automatic prompts point the child back to a word where something is wrong so that they locate the error themselves, which is the difference between practicing a correction and rehearsing a mistake for a week.

Behind the exercises sits 8,000+ words of curriculum vocabulary aligned to multiple German federal states, so what a child practices at home matches what school expects. A reward world, co-developed with children's author Carlo Meier, is what keeps a child returning to the fourth session of the week.

Teachers and parents use the Coach dashboard: per-child progress, which patterns the errors cluster in, time on task, printable reports and certificates. A teacher usually runs Grafari as a fixed practice slot or as targeted support for the children for whom the class list is not working; a parent runs it as a routine.

How has it been spreading?

Grafari began as Orthograph, the first product of the ETH Zurich spin-off Dybuster AG, released in 2007, and spread first through Swiss schools and specialist teachers – with the published research doing most of the persuading. The Annals of Dyslexia paper (2011) reported a 54% improvement in the rate of learning progress for the revised version over the earlier one, and two University of Zurich user studies found spelling errors falling by roughly 35% on average in trained children against about 5% in controls, with the second study reproducing the first. Grafari Phonics won the Worlddidac Award.

By the October 2021 acquisition, close to 200,000 children had used the two Dybuster applications – the spelling program and Calcularis – between them. It runs in English, German, French, Italian, Spanish, and Dutch, and is used beyond Switzerland in Germany, the United States, Canada, Brazil, and India. The 8,000+ word curriculum vocabulary aligned to multiple German federal states has been the main lever for spread into Germany, because it lets a school adopt the product without adapting its syllabus.

Since 2021 it has been distributed as Grafari within Constructor Tech (2,000+ partners and customers, 2.5M+ platform users), which added a school channel to the existing direct-to-parent motion.

How have you modified or added to your innovation?

The product has been revised against measured outcomes rather than feature requests. The Annals of Dyslexia study (2011) compared a revised version against the earlier one and found a 54% improvement in the rate of learning progress, with children training 15–20 minutes four times a week for 12 weeks – the revision was the thing being tested, not a by-product. Two University of Zurich user studies preceded and supported it, the second reproducing the first.

The curriculum vocabulary has grown to 8,000+ words aligned to multiple German federal states, which is what made adoption outside Switzerland practical, and the product now runs in English, German, French, Italian, Spanish, and Dutch. The reward world was developed with children's author Carlo Meier rather than in-house, on the view that the motivational layer needed the same seriousness as the pedagogical one.

Since the 2021 acquisition, the product has been rebranded from Orthograph to Grafari, extended with the Coach dashboard for teachers and parents, and integrated into the Constructor School platform.

If I want to try it, what should I do?

Start with the product page: constructor.tech/products/learning/grafari-o, and for schools the schools page at constructor.tech/products/learning/practice/grafari-o/schools.

If you are a parent: create an account, add a child profile, and let your child work unaided for the first few sessions while the system learns which patterns they get wrong. Then open Coach and look at the error patterns, not the score – the point of the product is that practice narrows onto those patterns.

If you are a school or a specialist teacher: begin with one group of children for whom the class spelling list is not working. Run 15–20 minute sessions four times a week for 12 weeks, which is the pattern used in the published studies. Check that the curriculum vocabulary matches your syllabus first – the 8,000+ word set is aligned to multiple German federal states, so confirm the fit for your region before you start. Review Coach weekly and compare progress against your own records at the end of the block.

To talk to us: get in touch through constructor.tech and ask for schools onboarding, which covers class account provisioning and a dashboard walkthrough.

Implementation steps

Check the curriculum vocabulary matches your syllabus
Before anything else, if you are a school, confirm that the 8,000+ word curriculum vocabulary fits your region. It is aligned to multiple German federal states, so the fit is excellent in some places and needs checking in others. A parent can skip this step; a school that skips it discovers the mismatch mid-term.
Set up child profiles
Create an account and add a profile per child with age and school year. No adult sets a difficulty level: the system selects words from each child's own error behavior, so it needs a short period of use before practice narrows onto that child's failing patterns. Schools should ask for schools onboarding and provision the group at once.
Let the child work unaided while the system learns their errors
Have the child run the first few sessions on their own. Automatic prompts point them back to a word where something is wrong so that they find the error themselves – an adult correcting over their shoulder removes exactly the step the product is built around.
Run 15–20 minute sessions four times a week for 12 weeks
This is the pattern used in the published studies, including the one reporting a 54% improvement in the rate of learning progress. Book it as a fixed slot rather than leaving it to available time. Twelve weeks is the shortest block from which you should expect to draw any conclusion.
Review Coach weekly, then evaluate the block against your own records
Look at which spelling patterns the errors cluster in rather than at totals that is what tells a teacher which children need something different, and it is not visible from a marked exercise book. Print reports and certificates as children progress. At the end of the twelve weeks, compare progress against your own records for comparable children before widening use.

Spread of the innovation

loading map...