Screening
We visit the plant together with your occupational safety and IT specialists: we examine the stock of cameras and access to the video stream, the critical areas and the priority risks. One working day.
Line of business
Digitalisation and artificial intelligence
The company’s second line of business is the practical adoption of artificial intelligence in industry. At its core is Estimatix: industrial safety video analytics that runs on the cameras already installed at the client’s site.
01 / Capabilities
What the system does
The system analyses the video stream from the plant’s existing cameras in real time and records breaches of occupational safety requirements. When a breach is detected, an event is logged with the date, time, location and a photograph, a notification is sent to the person responsible, and a summary report is compiled for the plant.
Hard hats — presence and fastened chin strap. Workwear, safety footwear, high-visibility vests and armbands. Safety goggles, gloves, face masks, hose respirators. Gas detectors and alarms. Fall arrest systems and carabiner attachment. Fire extinguishers and signs for gas-hazardous and hot work.
Position of personnel relative to the load and machinery. Danger zones, passage and movement beneath a suspended load. Slinging operations: conformity with the slinging diagram, skew, swinging, trajectory, idle time in suspension. Crane operation from the cab and by remote control. Monitoring of the audible warning during hazardous operations — by audio analytics, without separate equipment.
Personnel working at height and under the boom. Time spent in a confined space. Headcount within a zone. Open flame. Warning tape and cones. Self-diagnostics of the camera itself: lens contamination, lost focus, signal loss, camera displacement.
Monitoring of production stages and detection of defects. Warehouse logistics: loading and unloading operations, pallet accounting, vehicle movement routes, obstructed passageways. Site and entrance monitoring. Labour discipline and time tracking. Pre-shift checks of workers’ condition. Reports and notifications integrate with WMS, ERP, email and messengers.
02 / Demonstration
How it looks
Below are recordings of the system working at operating sites. The overlay in the frame is the system’s actual output; the footage has not been altered in any way.
03 / Technology
Why this has become possible only now
Classical computer vision systems required a separate model for each indicator and tens of thousands of labelled images to train every one of them. The barrier to entry was such that only a handful of plants could afford video analytics. The system is built on multimodal large language models — one model instead of seven, about a hundred video examples instead of tens of thousands of images, and the safety rules are set out in the text of the plant’s own regulations rather than in program code. When the requirements change, it is the text that changes, not the system.
Data per indicator
classical computer visionover 10,000 labelled images
the Estimatix projectabout 100 video examples
Models per task
classical computer vision7 or more
the Estimatix projectone
Defining the task
classical computer visionprogram code for each indicator
the Estimatix projectthe text of the regulations (PDF, DOC)
Change of requirements
classical computer visionretraining from scratch
the Estimatix projectediting the text of the requirements
Deployment time
classical computer visionlong months of debugging
the Estimatix project1–2 months
The barrier to entry has fallen by an order of magnitude.
Discuss a pilot at your plant04 / Sequence of work
How deployment is arranged
Findings
You receive a document: at which areas deployment makes sense, which monitoring markers work on the existing cameras, what will require additional equipment, the scope of work and the cost.
A pilot at one area
A limited set of markers, a short timeframe, a measurable result. It concludes with a report, hardware requirements and the terms of reference for the main project.
Roll-out across the plant
Extension to the remaining areas, integration with your systems, support and further training of the model during operation.
As the project is being launched on the Kazakhstani market, screening is carried out free of charge. What is required from the plant is one working day and the participation of occupational safety and IT specialists. The screening creates no obligations.
The offer applies during the market entry stage · August 2026
05 / Limits of applicability
Where the system has its limits
Not every camera is suitable. On low-resolution monochrome cameras, indicators that depend on colour, fine motor movement and precise distance do not work reliably: on such cameras the boundary between the load and the floor merges, and the margin of error in height reaches 30–50 cm against 2–3 cm on a colour camera with mapped planes. We set out such areas separately in the findings — stating exactly what additional equipment would deliver.
Requirements fall into four categories. What works on the current cameras straight away; what requires a camera to be replaced or moved; what requires data collection after additional equipment is installed; what requires a separate feasibility study. You receive this breakdown as a list against every item of your terms of reference — before the work begins, not after.
06 / Legal framework
We deploy within the legal framework of Kazakhstan
Video analytics at a plant means processing employees’ personal data, and it has to be launched on a documented legal basis, not afterwards. Before installation we check and put in order four things: the employees’ consent to video surveillance (standard HR consent forms at Kazakhstani plants as a rule contain no such clause — and without it any disciplinary action relying on a recording is vulnerable), the processing policy and the retention period for recordings, the appointment of a responsible officer, and the location of the storage within Kazakhstan — this is required by the personal data protection rules.
There are also functions that we deliberately leave out of the terms of reference even where they are technically available: facial recognition of visitors and audio recording in areas where information about a person’s health is spoken. Staff rest and dining areas are not included within the filming perimeter.
07 / Application
Industries
Railway engineering, rolling stock and wheel manufacturing. Oil and gas, and power generation. Metallurgy and metalworking. Mining and heavy engineering. Chemicals and pharmaceuticals. Food production, woodworking, textiles, building materials. Construction: technical supervision, materials accounting, acceptance. Warehouse logistics and distribution centres.
When it makes sense to start the conversation
the site has IP cameras and access to the video stream;
lifting, crane or other hazardous operations are carried out;
there are regular inspections and a history of occupational safety incidents;
the scale is several areas or more.
08 / Evidence
A case in railway engineering
A railway engineering plant: the wheelset and axle production area. Automatic monitoring of the use of personal protective equipment and of the safety of lifting operations by overhead cranes. Deployed on the existing monochrome cameras without replacing equipment — 21 monitoring markers in three months.
21 markers
of monitoring
3 months
deployment time
no equipment replaced
The name of the plant is not disclosed pending its consent.
09 / Deployments
Where the system is already running
An oil producer among the industry’s top ten and the principal employer of its town. Monitoring covers the working zones of four types of work: gas-hazardous, hot work, work at height, loading and unloading.
A metallurgical plant among the world’s ten largest copper producers. Deployment in melting shops at temperatures up to 250 °C: heat flows distort the image and produce swings in brightness — conditions in which ordinary video analytics stops distinguishing objects.
In Kazakhstan the line of business began with a pilot at a railway engineering plant — the case above.
10 / Transferability
The same mechanism in other industries
The system is not tied to industry: the set of monitored indicators is defined by text rather than by program code, so moving to another sector means changing the requirements, not developing from scratch. The two recordings below were made in catering and in retail.
11 / Line of business
Who leads the practice
Yerlan Dukembayev
Director of digitalisation at SABR
Director of the digitalisation department at the Bureau of National Statistics of the Agency for Strategic Planning and Reforms of the Republic of Kazakhstan (2024 — February 2026). Previously — head of the digitalisation project office at the Akimat of Pavlodar Region and director of the consulting department at an IT company.
Government information systems: the 112 and 109 unified emergency dispatch services, the registers of the Prosecutor General’s Office, the “Shekteu” unified debtor database. Over 15 years of experience, more than eight of them managing IT projects.
Responsible for defining the task, accepting the result and integrating the system into the plant’s processes.
12 / Contact
Discuss a pilot at your plant
We start by screening your areas — and on its results we issue findings on whether deployment is worthwhile.
WhatsApp: +7 701 817 6109
WeChat: olzhas_ryskeldi
Telephone: +7 701 817 6109
Astana · UTC+5 · Mon–Fri 09:00–19:00