Frequently asked questions

Glotech is an AI infrastructure and MLOps consultancy that takes AI systems from development into production. We work across AI architecture, MLOps, cloud infrastructure, DevOps, observability, computer vision and AI system integration.

We build the infrastructure that keeps your AI applications running in production. That includes model training and serving pipelines, cloud-native architectures, Kubernetes platforms, CI/CD, observability, computer vision systems, data pipelines and AI integrations.

No. Most of our work is on AWS, but we also work with Google Cloud and with cloud-agnostic tools such as Kubernetes, Terraform, Docker, Prometheus, Grafana and MLflow.

Both. Depending on the project, we work across the full lifecycle, from data and model development to deployment and production operations.

MLOps is the engineering discipline that keeps machine-learning systems reproducible, observable and maintainable in production. It applies software practices such as versioning, CI/CD, monitoring, experiment tracking and automated deployment to the ML lifecycle.

Yes. A project does not have to start from scratch. Depending on your system, we can improve its deployment, infrastructure, observability, scalability, model-serving architecture or ML lifecycle.

Yes. We build production computer vision systems: inference pipelines, data-centric training, edge and cloud deployment, model monitoring and human-in-the-loop review.

Yes. We design deployments around your existing network, cameras, applications and security requirements. If your systems cannot or should not be exposed to the internet, edge infrastructure runs inside your environment.

For general enquiries, contact info@glotech.io.

Kaya measures how many hours each machine actually worked by reading the machine's signal or tower light from camera footage. It does not attach sensors, switches, current clamps or any other hardware to the equipment. It needs no access to the machine controller, PLC, G-code or machine protocol. The only requirement is a tower light or signal lamp that a camera can see.

When your ERP or job data is connected, Kaya shows hours per customer. The tower light decides the hours. Which customer they belong to comes from the work orders in your ERP. You can print the report or export it to PDF or CSV.

Yes, as long as each machine has a tower light that a camera can see. Kaya works with any brand, age or type of machine. Nothing is installed on or inside the equipment, and machine warranties stay as they are. If you add a tower light to a machine that has none, Kaya can measure it.

Kaya provides the runtime and downtime data used to calculate Availability. It uses the factory's existing CCTV when the cameras can see the tower lights clearly. Where they cannot, additional cameras may be needed, or the camera angles may have to change. Performance needs part counts and an ideal cycle time, and Quality needs good and scrap counts.

Often, yes. If the recorder still holds the footage, Kaya can analyse past recordings. How far back it can go depends on the recorder's retention period.

Kaya is Glotech's own product. It measures how many hours each machine actually worked by reading the machine's signal or tower light from camera footage.

No. Kaya is a product we developed to measure machine runtime. It is offered separately from our consultancy work.

Kaya does not depend on machine make or industry. The only requirement is a tower light or signal lamp that a camera can see. It works on CNC machines, presses, moulding machines, packaging lines, assembly cells, robotic cells and other industrial equipment.

Kaya is for businesses that need to show their own management how many hours a machine ran. That includes contract manufacturers, companies that invoice by the hour, companies comparing hours across machines or sites, and businesses that use machine hours for cost accounting.

Kaya cannot measure equipment without a signal lamp that a camera can see. It is also the wrong tool if you need internal controller data, machine control, safety functions or a complete OEE calculation from Kaya alone.

No. Kaya provides the runtime and downtime data used to calculate Availability. Performance and Quality require additional data: Performance needs part counts and an ideal cycle time, and Quality needs good and scrap counts. Kaya can be integrated with OEE systems.

No. Kaya does not attach sensors, switches, current clamps or any other hardware to the equipment. It reads the tower light through cameras.

No. Kaya needs no access to the machine controller, PLC, G-code or machine protocol. It measures runtime from the tower light visible on the outside of the machine.

No. Nothing is installed on or inside the equipment. The machines are not opened, rewired, reconfigured or taken out of production, and operators keep working as usual. Installation can be done remotely.

Not necessarily. Kaya uses the factory's existing CCTV when the cameras can see the tower lights clearly. Where they cannot, additional cameras may be needed, or the camera angles may have to change.

It depends on your network and security rules. If the camera recorder can be reached securely from outside, Kaya connects directly to your existing recorder over an encrypted link. If your factory does not allow external access to its cameras or internal systems, Kaya uses a small edge device inside your factory network. The device runs AWS IoT Greengrass, communicates securely outbound with Kaya's AWS infrastructure, and transfers only the data needed for processing. It connects to the factory network but has no connection to or dependency on any production machine.

No. Cameras and the recorder can stay entirely private. Where external access is not allowed, Kaya uses a small edge device inside your factory network to connect securely outbound to Kaya's cloud services. This can even be a small Linux virtual server on your existing server system. The camera system never has to be exposed to the internet.

About an hour. You can set it up yourself. Once Kaya is connected to your own AI assistant, it guides you through each step. Or the Glotech team can manage the onboarding with you. If a small device has to go on your factory network, the time stays the same as long as you have a Linux server ready.

No. Setup can be done entirely remotely.

Setup covers the company and site, the cameras, the machine list and which camera sees which machine. Operator information is added where relevant. In each camera view, a zone is drawn around every machine's tower light and, where needed, around the operator or motion area.

You redraw the zones in the zone editor. Every mapping is saved with a revision number and timestamp. A change never overwrites the earlier configuration.

Yes. Kaya uses AWS IoT Greengrass where processing or data access has to happen inside the factory network. The device communicates securely with Kaya's AWS infrastructure.

Yes. Glotech does not need to host your ERP. Kaya can connect to an ERP in your own environment, in line with your network and security rules. Where direct external access is not allowed, an edge device retrieves the ERP data Kaya needs and transfers it securely. The ERP supplies context such as work orders, machine assignments, operators, parts and customers.

No. Kaya reads only the ERP data needed for the configured integration and reports. Runtime itself is measured from the tower light.

No, it is optional. Kaya measures runtime without it. An ERP connection adds work orders, machines, operators, parts and customers, so you can assign hours to jobs and customers.

No. The tower light is the only source of runtime. ERP data adds context to the measured hours but cannot add, remove or change them.

Yes. Cameras and the recorder can stay inside your network, closed to the internet. If the recorder can only be reached from inside the factory, a small computer on your network connects securely to Kaya's cloud services. The exact setup depends on your network rules, which we review with you before installation.

No. Where a customer does not allow inbound access to factory systems, Kaya runs through an edge device inside the customer's network.

Kaya reads the machine's signal or tower light and, when the light is amber, checks whether anyone is at the machine. It reports running, stopped, alarm, nobody at the machine (amber light), light off and no-evidence periods separately.

Only when the light is green. Green counts as running whether or not anyone is nearby. When the light is amber and nobody is at the machine, Kaya reports that time separately and excludes it from measured runtime. Customers can use the resulting hours for billing or internal reporting.

Kaya does not guess. If a camera is down, the lamp is blocked or the image is not clear enough, that period is recorded as no evidence. It is never given a machine state.

Kaya does not assume the machine kept running. Periods without enough evidence are reported as no evidence, so zero runtime is never confused with unmeasured time.

Yes. Reported hours link back to the underlying camera frames and state transitions. You see the signal or tower light state, detected presence, neighbouring frames and how the state changed.

Often, yes. If the recorder still holds the footage, Kaya can analyse past recordings. How far back it can go depends on the recorder's retention period.

Part and cycle counting is an optional capability for specific setups where parts move through a defined camera zone. Counting is completely separate from runtime measurement and never changes measured hours. Where part counting is not configured, Kaya simply marks that data as not available rather than estimating.

No. Kaya only checks whether a defined area at the machine is occupied. It does not identify individuals, follow people between machines or produce performance figures for named people.

No. Kaya measures and reports what can be seen from outside the machine. It complements PLCs, MES, SCADA and machine controllers rather than replacing them.

Results depend on the equipment, the tower lights, camera coverage, image quality and conditions on the floor. You can open the camera frames behind any hour and check it yourself.

Kaya provides floor, day, and machine views. When connected to your ERP or job data, it also provides customer views, work-order coverage, and hours per customer. You see running hours, stopped time, time with the light amber and nobody at the machine, and machine sessions. Print reports or export them to PDF or CSV.

Yes. Kaya connects to your own AI assistant. The assistant can help with onboarding, look up machine and site information, answer questions about runtime and open the evidence behind reported figures.

For example: How many hours did this machine run last month? Which machines spent the most time on amber with nobody there? What happened on the shop floor yesterday? Which days had the most impact on this machine's runtime? How many hours did we run for this customer? Show me the footage behind these hours.

Kaya runs on AWS with encryption in transit and at rest, role-based access control, tenant isolation and audit logging. Greengrass edge devices are authenticated with a unique X.509 identity per device and mutual TLS.

We prepare an offer after reviewing your site, your cameras and what you want to measure. The first 30 days are free.

Yes. You can try Kaya free for 30 days. It is installed on your existing cameras and measures how long each machine runs. At the end, we go through the results with you. If your cameras do not see the tower lights well enough, we look at that with you first.

First, we check whether your machines have tower lights that your cameras can see. Then we look at camera coverage, how long your recorder keeps footage and whether it can be reached from outside. We also ask whether you want hours linked to jobs in your ERP.

Contact info@glotech.io.