One AI, Millions of Behavioural Patterns. And the Potential to Transform Entire Industries
www.posiv.cz ǀ 53
STARTUP
How does GPRT work? What can users learn about
their gameplay?
We have two models – GPRT Flash and GPRT Pro –
which can predict in-game events. For example, with
over 90% accuracy they can estimate who will win
a particular round in a shooter like Counter-Strike. They
also predict who is likely to survive, who will complete
an objective, or which path a player will choose. And it
is precisely on these predictions that we want to build
the next features of our platform.
There are plenty of numbers, but visualisation alone
isn’t enough. We wanted to “translate” data into
clear, actionable feedback that players can actively
work with. And that is the main goal of our platform
– to make analysis understandable, easy to navigate
and practically usable.
Today we have analysed more than 6 million players
– representing thousands of years of gameplay. That’s
a scale of experience that no individual could ever
imagine. When someone has such a “digital coach”
at their side, telling them based on patterns: Here
is probably the best decision, they can make choices
based on data – not just intuition. And I believe this
is the direction society as a whole is gradually heading.
It does sound like the volume of data you’re working
with is enormous.
It is. At the moment we’re handling roughly half
a petabyte of data – around 500 terabytes. That’s
an unprecedented scale, which also brought very
specific technical challenges.
For example, when training the models, it quickly became
clear that downloading such an amount over the internet
would take weeks and cost a fortune. That’s why Amazon
sent us special data containers – essentially portable
servers – directly from Frankfurt to Ostrava. We were able
to transfer the data from them via cable within two days.
It’s an experience very few AI start-ups ever go through.
How do you work with the data you’ve collected?
What has it brought you?
We’ve proved that we can clean, structure and prepare
data efficiently for AI training. We also classify data
quality into three levels – gold, silver and bronze.
We use modern AI architectures such as transformers
and cross-attention mechanisms, and we continuously
monitor global developments to make sure we’re
applying the latest capabilities.
Your AI models are currently focused on the gaming
environment. Do you have ambitions to go further?
Yes, the technology we’ve developed has universal
applications. Gaming data allowed us to train
the models effectively, but the principle of a decision
support system can be transferred to other fields –
from education and healthcare to defence and public
administration.
For example, in education, if we obtain data on student
performance, we can identify which forms of testing
lead to improved results. In healthcare, AI can detect
whether a particular treatment method is slower,
and uncover patterns between diagnoses, treatments
and risk factors that may escape human observation.
And in gaming we simulate stress and decision-making
– just like in crisis or military simulations, where we see
major potential for defence.
In our region, heavy industry dominates, and this
is precisely where we see huge potential for AI
applications. We can create a system that monitors
the condition of machines and equipment in real time
and predicts when maintenance or part replacement
is needed – before a breakdown occurs. Through visual
analysis it can also detect wear or other indicators that
in the past have led to failures.
Do you already have concrete discussions beyond
the private sector?
Yes, we collaborate with the City of Ostrava,
the Moravian-Silesian Region and with the Černá
Kostka institution, which brings technology closer
to the public. If we have access to relevant data –
for example on traffic or air quality – we can build
predictive systems to enable more efficient city
management.
At the end of the day, it’s always a human who makes
the decision; however, the better the information they
have, the better that decision will be.
Thank you for the interview.
Game data helped train the models,
but the system’s principle can be applied
to other areas as well.