Offline Face-Recognition Attendance: How Hazeri Works

Still tracking attendance in a paper book or a group chat? An old phone at the door can do it with no internet.

Face-recognition attendance means a camera compares each person’s face with enrolled samples and records their check-in and check-out. In Hazeri everything runs on the phone itself, with no internet and no stored photo; only a numeric vector is kept. It suits a small business, not bank-grade security.

By Saman Ghahreman · Published: · 7 min read

Case studies and build experience

Offline Face-Recognition Attendance: How Hazeri Works

In most small workshops, clinics, cafés and shops, attendance is still recorded in a paper book or a chat group. Hazeri is another way: put an old phone or tablet at the door and it becomes your attendance device. This article explains how it works, what it cannot do and what you must take care of.

Why is attendance still a problem?

Every method has a weakness. A colleague can sign or swipe a card for someone else. Dedicated face-recognition devices usually cost more. Cloud software needs a stable internet connection and a subscription, and it puts staff data on another server.

MethodStrengthMain weakness
Paper book or chat groupSimple and freeEasy to fake and reports are manual
Card or fingerprint readerFastA card can be handed to someone else; a separate device is needed
Cloud softwareOne report and many branchesDepends on internet and a subscription; data sits on a server
HazeriOn one old phone, no internet and no stored photoOne independent device; syncing several branches needs a custom version
The path of one check-in in Hazeri
  1. Look at the camera

    Each person stands in front of the phone or tablet at the door.

  2. Face recognition on the device

    The face is compared with the enrolled samples; nothing goes to a server.

  3. Liveness challenge

    Blink or turn the head, so a photo or video is not accepted.

  4. Check-in or check-out recorded

    The time is saved and appears in the daily and monthly Persian-calendar reports.

How is a face recognised?

At enrolment, five face poses (front, left, right, up, down) are captured for each person and the model builds a numeric vector from each. Later, the face in front of the camera is compared with these vectors; if the similarity is above the configured threshold and nobody else scores close to it, the identity is confirmed. No face photo is stored. The model is the open-source Human library, which runs on the device itself.

What if someone shows a colleague’s photo?

That is why it has a three-layer liveness check: the model’s anti-spoof score, a random challenge (blink or turn the head) and identity stability over several consecutive frames. This is enough for a small workshop, clinic or shop, but it is not bank-grade security. If your site is more sensitive, raise the liveness level and add a human check.

Why offline and with no internet permission?

  • In many places the internet is unstable; attendance should not stop when it drops.
  • The Android app has no internet permission, so it cannot send data itself. Anyone can check this in the app’s permissions.
  • Data stays in the app’s private storage on that device (in the web version, in the browser’s internal database called IndexedDB; the web version also works offline after the first load).
  • No server means no server subscription cost; in return each device is independent.

Setting it up in a workshop or clinic

  1. Put an old phone or tablet running Android 6 or later at the door and give it power and a fixed spot.
  2. The first time, create an admin PIN and, if you like, turn on fingerprint unlock.
  3. Enrol each person with a name, a code and five face poses (in soft light).
  4. Set shift start and end, the lateness grace and the working days.
  5. Try it with staff for a few days; if someone is not recognised or two people are confused, adjust the similarity threshold in settings.

Staff privacy: what must you take care of?

A face is biometric data and one of the most sensitive kinds of staff data. The technology is only a tool here; the responsibility is your business’s.

  • Before starting, tell staff the purpose and how it works and get their consent.
  • Keep only the data you need, and remove anyone who leaves; deleting them also erases their face and records.
  • Turn on the device screen lock and the admin PIN; the data inside the app is not separately encrypted.
  • Keep backup and Excel files somewhere safe.
  • Check your country’s and trade’s rules with a specialist. The NIST Privacy Framework and responsible-use rules for AI are a good start.

Full text: Hazeri privacy policy.

Where is it not suitable?

  • Very high-security places (banks and sensitive sites): Hazeri’s liveness check is for small businesses.
  • Several branches with one combined report: each device is independent and syncing needs a custom version.
  • A payroll link or alerts to the manager: not in the current version and can be built as a custom job.
  • A very large staff: Hazeri is designed for about eighty people.

If you run a training center, you can lighten replies to parents next to attendance with a training-center chatbot. And if you need a custom AI model, read custom AI models: prompting, RAG or fine-tuning.

Frequently asked questions

Is face-recognition attendance possible without internet?

Yes, if the model runs on the device itself. Hazeri works this way: the models are inside the app and the internet permission is removed from the Android app.

Can the system be fooled with a photo or video?

Hazeri has three liveness layers (anti-spoof score, random challenge, identity stability), which is enough for a small business. It is not bank-grade, though; sensitive sites also need a human check.

Do I need staff consent?

Yes. A face is biometric data; inform staff, get their consent and check your country’s and trade’s rules with a specialist.

Where can I try Hazeri?

Download the free demo from the Hazeri page. For setup in your business or a custom version, order from the contact page; you see it working before you pay.

Sources

Tags: face recognition attendance · offline attendance · attendance app · on-device face recognition · Hazeri

Get advice on Custom AI Model Development

One short consultation is enough to clarify the path, timeline and rough cost of your project.

My promise
If you have a problem you think AI might solve, write to me. I’ll tell you honestly whether it will; sometimes the answer is no, and hearing that early is cheaper.
Book a consultation