Nazarbayev University — advanced multi-module testing at scale

The situation

Nazarbayev University was the deployment where the question stopped being whether the platform worked and became how far it would go.

Building on the methodology established at the earlier pilots, the team deployed a significantly expanded module set across twelve campus buildings — making this the most comprehensive test environment AMAN has run to date.

Seven modules — the full stack

The core detection suite was extended with four additional modules.

The core suite, as deployed at every site:

• Weapon detection — real-time alerts when firearms, knives or hazardous objects appear on camera.
• Smoking detection — automated detection of smoking in restricted zones across the campus.
• Aggressive behaviour detection — pattern recognition for physical altercations and repeated conflict behaviour.

Extended at NU:

• Face recognition — identity verification and person-of-interest detection across campus entry points and high-traffic zones.
• Camera sabotage detection — immediate alerts when a camera is obscured, repositioned or physically interfered with.
• Fall detection — automated detection of falls, which covers medical and welfare incidents as well as safety hazards.
• Abandoned object detection — flags unattended bags, packages or items left in public spaces beyond a defined period.

The four extended modules are not part of a standard schools configuration. They were deployed here because a multi-building university campus is where a module set this size can be evaluated properly, and having them running is the reason this pilot produced the results it did.

Rigorous testing at high traffic, high complexity

A larger module set demands proportionally harder evaluation, not the same evaluation repeated seven times. Each module was tested across a range of real campus scenarios:

• Detection quality. Each of the seven modules assessed for accuracy in live university environments — lecture halls, open spaces, corridors and entry points.
• Response accuracy. Alert precision tested to confirm the system flags genuine incidents without burying security staff in false positives.
• Edge case handling. The difficult scenarios, deliberately: unusual angles, occluded views, crowded scenes and low light. These are where detection systems quietly fail, and where a demonstration would never have found the limits.

What changed

NU produced the most comprehensive results of the three deployments.

• The richest dataset of the three — broadest module coverage, highest traffic volume
• The platform validated at scale across a multi-zone, high-traffic campus without performance degradation
• Scalability confirmed from a single-school setup through to full campus deployment
• Edge case handling evaluated across all seven modules in real conditions
• Results and recommendations delivered to university leadership

Read alongside Rudny and MNU, this is the argument the three studies make together: the same platform, deployed on the same principle — audit first, existing cameras, no hardware replacement — scales from one school building to a twelve-building campus without becoming a different product.