A Grade 2 teacher in Riyadh has thirty children and forty minutes. If she gives every child a fair turn at reading aloud, each one speaks for about eighty seconds. That is before she has explained anything, settled the room, or corrected a single vowel.So it doesn't happen. Most primary children read to their teacher for a minute or two a week. The rest of their speaking goes unheard, and whatever they get wrong, they keep repeating until the habit sets.That arithmetic is the reason spoken English lags behind written English in so many primary schools. It isn't a curriculum failure or a teaching failure. Pronunciation is learned by speaking, being heard, and being corrected. Speaking scales fine in a classroom of thirty. Being heard and corrected does not.
This is the exact gap ai english learning was supposed to close, and for a long time it didn't, because most of the tools were designed for adults with headphones, quiet rooms and the patience to self-study. Seven-year-olds sharing a school tablet are a different problem entirely.ZAIRA was built for the second group.
Why pronunciation is the hardest part of English to teach at scale
Arabic and English don't share a sound system, and the mismatches are predictable enough that any experienced teacher can list them.There is no /p/ in Arabic, so pen drifts toward ben and copy toward coby. There is no /v/, so very comes out as fery. English allows consonant clusters at the start of a word that Arabic does not, so street picks up a helping vowel and lands somewhere near istreet. And a language built on three core vowel qualities has to stretch to cover the dozen or so English uses, which is why ship and sheep collapse into the same word.None of this reflects effort or ability. It's motor habit. A child who says /b/ for /p/ four hundred times in Grade 1 will not unlearn it in Grade 7 through worksheets. It has to be caught early, by ear, in the moment it happens — which is a one-to-one job, and one-to-one is the thing a full classroom cannot supply. It is also why generic ai english learning apps built for adult self-study tend to fall apart the moment they meet a kindergarten class.
What ZAIRA actually does
ZAIRA is an ai english learning platform built for students from kindergarten through Grade 6 in Saudi schools. The core is simple to describe and difficult to build: every child gets an AI listener that hears them read, evaluates how they pronounced it, and answers immediately with feedback and another attempt. As many attempts as they want. In class or at home. Nobody waits for a turn.
Around that core sits a full school system rather than a standalone app:
A student experience on web and mobile, designed so the same account and the same progress follow the child between school and home.A teacher portal for assigning practice, tracking a class, and spotting who needs direct attention.A school administration portal for managing teachers, classes and enrolment, with visibility across the whole school.Parent access, with controls over how their child uses it and a clear view of what they've been practising.
Four audiences, four genuinely different views. Building one interface and exposing it four ways would have failed all four.
Children's voices break most speech engines
Most speech technology is trained on adult speakers. Children are higher in pitch, irregular in pace, inconsistent in volume, and still incomplete in articulation. Systems tuned for adults degrade sharply on them.
There's a harder problem underneath that. In dictation, a mispronunciation is noise to be filtered out. Here it is the entire signal. The system has to hear the mistake clearly and then decide something a transcription engine never has to decide: is this an error, or is this a five-year-old whose mouth hasn't got there yet?
Some sounds simply are not acquired at that age. Flagging them would be both technically wrong and quietly discouraging. So expectations shift across the age range. What counts as a miss for a Grade 5 student is normal development in KG, and the feedback is written to reflect that.
The engineering behind this sits close to the work our team does on AI Voice Agent & Voice AI Solutions, where accent, intent and noisy real-world audio are daily problems rather than edge cases.
Confidence is the real bottleneck
Ask any primary teacher which children need reading practice most, and she'll name the ones who never volunteer. Speaking in front of classmates is a social risk. Children who suspect they'll get it wrong avoid the risk, and avoidance is what turns a small gap into a large one by Grade 6.
An AI listener removes the audience. There's no laughing, no sighing, no next child waiting. So children try words out loud that they would never attempt in front of the class, and they try them ten times instead of once.
We built the loop around retry rather than judgement, and wrote the feedback to keep a child going rather than to mark them wrong. A tool that makes a seven-year-old feel stupid is a tool they stop opening, and after that no amount of accuracy in the assessment matters.
What teachers and school leaders actually get out of it
The point of ai english learning in a school is not to replace the teacher. It's to tell her where to spend the little one-to-one time she has.
Judging pronunciation by ear leaves no record. Standards drift between teachers and across a school year, and a teacher who knows a child is struggling often can't show a parent precisely with what. Every ZAIRA session produces data, so the class view shows who is on track and who needs help — not a raw dump, but the two or three names worth acting on this week.
Over a term, that turns into something schools have rarely had at primary level: evidence. Improvement you can demonstrate to a parent, compare across cohorts, and report on. Better English learning outcomes stop being an impression and become a number.
One rule governed the whole design. Teachers are already stretched, and any platform that adds marking, admin or troubleshooting gets abandoned no matter how good the technology is. If it doesn't visibly reduce workload in the first fortnight, it has failed.
Parents, and the practice that happens at home
English learning doesn't stop at the school gate, and for pronunciation the home hours matter more than most subjects, because that is where repetition is possible without an audience.
Parents can see what their child has practised and manage their access. A child works on mobile at home in the evening, and the teacher sees that practice reflected in the class record the next morning. Same account, same progress, both settings.
Data handling was treated as a design constraint rather than a policy page. A platform holding voice recordings and performance records for primary-age children carries the highest duty of care there is. Role-based access is enforced by the system: teachers see their own classes, administrators see their own school, parents see only their own child.
What Saudi schools should look for in an ai english learning platform
If you're evaluating options this academic year, the questions that separate a real school platform from a consumer app are fairly blunt:
Was it built for children's voices, or adapted from an adult product? Ask specifically how it handles KG speech.Can a pre-reader operate it alone? If a child needs an adult to read the screen, it will only ever be used under supervision.Does it distinguish developmental stage from error? Age-banded expectations, not one standard across seven year groups.Does it survive a real classroom? Shared tablets, background noise, variable connectivity, short periods.Who can see a child's data? Role boundaries should be enforced in code, not by trust.Does it reduce teacher workload? Adoption dies here more often than anywhere else.Does it run on web and mobile? Practice has to follow the child home.
Where this sits in Saudi Arabia's AI education push
English is taught from the early primary grades across Saudi schools, and spoken fluency has moved up the agenda as the Kingdom's economy has opened further to international business, tourism and technology. At the same time, national investment in AI education has made schools far more willing to trial genuine classroom technology rather than digitised worksheets.The two lines meet neatly here. Individual speaking practice is the one part of English learning that has never scaled — and it happens to be a problem AI is unusually well suited to solve. Used properly, tools like this don't hand the class over to a machine. They hand the teacher back the hours she was never able to find.
Frequently asked questions
Kindergarten. The interface is carried through visuals and audio instead of text, so a child who can’t yet read is able to use it without an adult driving the screen.
No. It handles the repetition, and it tells the teacher exactly where her attention is needed. The teaching stays human.
It was built for shared devices, noisy rooms and short lesson periods, and it runs on both web and mobile. Classroom conditions were the design brief, not an afterthought.
Building it, and getting it adopted
ZAIRA was engineered by Richinnovations from Technopark, Trivandrum, for a Saudi client. Eleven years in, we've learned that a school platform doesn't fail on the code. It fails when nobody in the building knows how to explain it to a principal, which is why the same team that ships the product also runs our Digital Marketing Company work — the positioning, the search visibility and the campaigns that put a good product in front of the schools that need it.
If you're planning an ai english learning rollout, a pronunciation programme, or any AI education project across the Gulf, we're happy to walk you through what worked in ZAIRA and what we'd do differently. That conversation is usually where a pilot turns into infrastructure.