AI · EdTech · Attendance

Rollio

A facial recognition attendance system for schools. A student simply looks at the tablet and is marked present in under a second, with recognition running on the device, fully offline, and a private signature stored instead of a photo.

Platform Web, Android & iOS

Category AI / EdTech SaaS

Recognition On device, offline

Privacy Consent first

The Project

Taking attendance by hand eats into the start of every school day, and the usual fixes bring their own problems. Card taps get shared, fingerprint readers worry parents, and cloud face systems send children's photos to servers no one in the school controls. Rollio was built to remove the daily friction without any of that.

The idea is simple from the front. A student looks at a tablet by the door and is marked present in under a second. The hard part sits underneath. Recognition runs entirely on the device, every scan works without the internet, and the system stores a private numeric signature rather than a photo. We built Rollio end to end, from the multi tenant web platform that onboards schools and assigns their devices, down to the on device model on the tablet, so a school can trust it with a child's face.

Under 1s

To recognise a student and mark them present, with no tap and no effort from the child

Offline

Every scan runs on the device, so the morning register works without the internet

No photos

A private numeric signature is stored for matching, never a child's photo

Rollio in Action

A calm, friendly moment for students at the door, a clear workflow for staff, and a web platform for onboarding and management

The kiosk recognises a student from a live preview, confirms a real person is present, and marks them present in under a second

On the web, a super admin onboards schools and assigns devices, and each school signs in to manage its own teachers, classes, and students

One app, two modes. A fixed kiosk for the door, and a staff mode for teachers and administrators

Staff can also mark a student by face from their own phone, using the camera or a gallery photo

Built to work offline and sync when it can, with English and Arabic supported from the start

The Challenge

Recognise a child in a second, and never get it wrong

Attendance for schools sounds like a solved problem until you look closely. It has to be fast enough that a line of students clears in seconds, accurate enough that no child is ever marked as the wrong person, and private enough that parents and regulators are comfortable with a system that handles children's faces.

Those goals pull against each other. Speed usually means guessing, accuracy usually means a server, and privacy usually means more taps and more friction. The brief was to hold all three at once, on the kind of ordinary tablets a school already owns.

Because every scan touches a child's face, we treated consent and privacy as the foundation rather than a feature added at the end. A face is held and never used for recognition until a verified parental consent record exists, and withdrawing consent stops recognition straight away.

⚡

Look and go

A live preview captures a clear, frontal face automatically, so marking attendance needs no tap and no effort from the student.

🛡️

Refuses to guess

A confidence threshold and a margin check between the top candidates mean a wrong student is never marked present. When unsure, the system asks rather than guesses.

🙂

Liveness built in

A passive anti spoofing check confirms a real person is present, so a printed photo or a screen replay cannot mark attendance.

🤝

Never blocks a child

When recognition is unsure, the teacher picks from the top candidates, and a help action is always on screen, framed as normal and never as a fault.

🔒

Consent first

A face goes live only after verifiable parental consent, and removing consent returns it to a held state and stops recognition at the next sync.

📶

Works offline

Recognition runs on the device with no network call in the common path, then syncs the register whenever a connection is available.

✍️

Mark by hand

When a scan is not the right fit, a teacher marks a student present, absent, or late by hand, and every manual change is recorded with who made it and when.

📅

Weekly insights

A weekly view shows attendance by day and flags repeated absence, so a teacher can follow up with the students who need it instead of digging through registers.

What We Built

A multi tenant web platform, two clear apps, and on device recognition

Rollio is built as a multi tenant platform. From the web, a super admin onboards a school and assigns the tablets and phones it will use. Each school then signs in to its own space to add its teachers, classes, and students, so the people closest to the children manage their own roster and their own data.

On the devices, Rollio is one app that runs in two modes. Kiosk mode turns a fixed tablet at the door into the attendance station, with a live preview, automatic capture, and a calm confirmation by name and photo. Staff mode gives teachers and administrators the class register, a mark by face option, a manual mark by hand option for when a scan is not the right fit, consent management, and weekly views.

Under the surface, every signature is built by a single recognition model, so all signatures stay comparable whether they come from a bulk photo upload, a spreadsheet of students and images, a student information system connector, or a capture taken on the tablet. The match runs on the device against the live part of the school gallery, takes its decision across several frames to remove random error, and only accepts a result that clears the threshold and clearly beats the next candidate.

Around recognition sits the part schools actually live in day to day. A consent dashboard tracks who is collected, verified, and outstanding. A live class dashboard shows present, absent and pending counts with arrival times. An offline first sync means a patchy connection never stops the morning register, and English and Arabic are supported from the start.

Tech Approach

On device intelligence, privacy by design

The core decision was to keep recognition on the device. A captured face is turned into a numeric signature and compared against the school gallery locally, so the common path needs no server, the experience stays fast, and a child's photo never leaves the tablet. What is stored for matching is a signature, not an image.

We built the apps cross platform for Android and iOS so a school can use the tablets and phones it already owns. A passive liveness model guards the recognition step, a shared embedding model keeps every signature comparable across sources and model versions, and the consent and isolation rules are enforced in the data layer, so a held face simply cannot be matched.

The platform itself is multi tenant. The super admin onboards schools and assigns devices from the web, and each school manages its own teachers, classes, students, and consent records in an isolated space, so one school can never see another's data.

Web App Multi Tenant SaaS Android iOS On device ML Face Recognition Liveness Detection Offline First Sync

Built for Trust

A system a school and a parent can trust

The whole flow holds together. A student is enrolled from existing records or on the tablet, a consent record is collected and verified before any face is ever used, and a student walks up to the kiosk and is marked present in under a second, offline, with the system refusing to guess when it is not sure.

Because consent, liveness, and on device matching were treated as foundations rather than additions, the harder questions a school and a regulator ask are answered by the way Rollio is built, not patched on at the end.

Under 1s

From looking at the tablet to being marked present, with no tap required

Offline

Recognition runs on the device, so attendance keeps working without the internet

No photo

A private numeric signature is stored and matched, never a child's image