You already have the cameras. Put them to full use.
Footage gets played back once something has happened. Computer vision reads it as it comes and tells you while the problem can still be dealt with: how many spaces are free, which part is scrap, who walked into the zone without a helmet.
Why choose me?
What you get out of working together, beyond a working solution
Works with the cameras you have
In most cases the existing IP cameras and their stream are enough. I only bring in new ones where the resolution or the angle genuinely will not carry the detail in question, and I can tell from a sample recording before quoting, not after signing.
The video stays with you
Recognition runs on a box on your network or in your own tenant. What leaves is a number or an event, not the footage. That removes most of the GDPR questions along with the bandwidth bill you would otherwise be paying.
Reacts now, not in a monthly report
Evaluation runs continuously, so the output can be an immediate alert: the queue is over the limit, that is the third reject in a row, someone entered the zone without PPE. Acting on it makes sense while there is still time.
Numbers nobody is collecting today
Occupancy over time, defect rates by shift and machine, dwell time at the loading bay. Data you cannot get at today because somebody would have to count it by hand. You can actually manage on it.
The legal side settled upfront
The AI Act treats workplace monitoring as high risk and bans biometric categorisation of people at work. I therefore build these systems to count and detect, not to identify. What is allowed and what is not comes to you in writing before anything gets built.
Output into your systems
A detection on its own is worth nothing. I wire it into what you use: a write into ERP or MES, a tile in Power BI, a message into Teams or email, a stack light on the line.
Types of solutions
Choose the solution that best fits your needs
Counting and traffic
How many arrived, how many are parked and how long the wait is.
- Car park occupancy and free spaces in real time
- Queue length at a counter or barrier
- Footfall counts and space utilisation over time
- Licence plate reading at entry and exit
Quality control
A defect caught on the line instead of at the customer.
- Detection of surface, shape and colour defects
- Completeness check before packing
- Label and code legibility and correctness
- Reject statistics by shift and machine
Safety and operations
An alert while nothing has happened yet.
- PPE checks: helmet, vest, goggles
- Someone entering a hazardous zone or a machine path
- Unsafe handling and blocked escape routes
- Anonymous evaluation, without identifying individuals
Warehouse and logistics
A real picture of what is standing where.
- Loading bay occupancy and dwell time
- Pallet counts and free rack space
- Loading checked against the delivery note
- Detection of empty and damaged pallets
Where to find me and where I work from
I am based in Prague and in Tálín near Písek. Consultations run online or in person. I work with clients across the Czech Republic and abroad.
Prague
Letňany and surroundings
Tálín near Písek
South Bohemia
Online
Czech Republic and worldwide
Frequently asked questions
Answers to the questions clients ask most often
Does it have to run in the cloud?
It does not, and usually it should not. The standard deployment is a box on your network processing the camera streams on site. Only the result leaves it, meaning a number or an event. The cloud earns its place once you want data from several sites together.
Do we need new cameras for this?
Usually not. What gets evaluated is an ordinary RTSP stream, so the cameras you already have will do. I raise new ones where the angle is wrong, the light is poor or the resolution will not carry the detail in question. I can tell from a sample recording before quoting.
How accurate is it, and what about false alarms?
Accuracy varies from task to task. Counting cars in a car park is reliable almost immediately, a fine defect on a glossy surface needs training on your data. That is why we start with a pilot, where accuracy is measured on the real operation and the decision to deploy follows that figure. False alarms get tuned with thresholds and rules, not with promises.
What does the hardware cost, and what does running it cost?
The hardware is a one-off in the order of tens of thousands of crowns per site, depending on the number of cameras and how demanding the task is. Running costs are low, because the model runs on your premises and there is no per-request fee to somebody else's API. You get the actual figures after the pilot, not as an estimate over the phone.
How long does a pilot take?
Weeks rather than months, on one line or one car park. The point of a pilot is not a finished system but measured accuracy and a documented saving you can take to the board. The pilot has a fixed price, and its conclusion may well be that going further is not worth it.
Are we allowed to film employees like this?
Under certain conditions yes, but it is a sensitive point and worth settling at the start. The AI Act places workplace monitoring among high-risk uses and prohibits biometric categorisation of people at work. I therefore build these systems to be anonymous: they detect that somebody in the zone has no helmet, not that it is Mr Novák. Scope, retention and informing staff get handled upfront, not during an inspection. I am not a lawyer, so I do not supply a legal opinion, but I will tell you what has to be evidenced from a technical standpoint and what is better left unbuilt.
What if the product or the lighting changes?
The model gets retrained on new data. That is why monitoring detection quality and topping up samples is part of the deployment, rather than a one-off handover and goodbye. Where the product mix changes often, I cover it with a maintenance retainer.
Do you need our footage for training?
For simple tasks no, a pre-trained model is enough. For quality control yes, because nobody else has your defects. The footage stays yours, it is used only for your model, and it is not sent off to train anybody else's. The trained model is yours.
Want to know what your cameras could be telling you?
Send a few minutes of footage and say what matters to you at that spot. Free of charge and without obligation, I will tell you whether it can be recognised reliably from that image at all and what a pilot would involve. Online or in person in Prague and South Bohemia.
Before I build anything, we go through your processes and I write up where AI pays off, how much it saves and what running it will cost. The audit fee is deducted from the project that follows.
Or go back to the contact form