VAELLA joins the NVIDIA Inception Program

Why we joined NVIDIA Inception

VAELLA builds Vision AI for crowded spaces: metro stations, airports, ferry terminals, offices and shopping centers. Our job is to see how people move through a building, and to help the teams running these buildings act on that in real time.

That only works if the AI runs real-time, inside the space where the people are. A station platform cannot wait for data to travel to a distant data center and back, and many operators do not want it to leave the building at all. Our product runs on three selected NVIDIA technologies that support this mission.

NVIDIA Inception is a program that supports startups building with AI. The membership connects our team more closely to the NVIDIA platform we already build on.

3 NVIDIA technologies used by VAELLA

Vision AI model training with NVIDIA Omniverse simulations

Collecting real footage of crowds is slow and raises privacy questions. We use NVIDIA Omniverse to create realistic, simulated scenes, such as busy platforms, lobbies and escalators, and train our Vision AI models on that synthetic data. The models learn about rare situations, like a platform filling up during a disruption, or a person falling, without waiting for one to happen.

Embedded edge AI with NVIDIA Jetson

Our Vision AI sensors use lidar technology, which measures distance rather than capturing images. A compact NVIDIA Jetson computer installed in the building turns those measurements into people counts, movement and dwell times on the spot. As an example, this setup can tell the elevators when demand is building in the lobby before passengers press a button.

Local inference with NVIDIA DGX Spark

VAELLA’s Building AI Operator lets anyone on the team ask about their building in everyday language. We run the large language models behind it locally on NVIDIA DGX Spark, so building data can stay on site. Where a customer chooses, the operator can also use frontier models in our EU cloud.

What this makes possible for our customers

"We train our Vision AI on simulated data in NVIDIA Omniverse, run them on NVIDIA Jetson inside our customers' buildings, and run our language models locally on NVIDIA DGX Spark.

For customers, the outcome is significant: more of the building measured for the same budget, faster deployments, and an operator that acts while a situation is still unfolding, not in next month's report."

— Jussi-Pekka Partanen, Co-founder & CTO, VAELLA

In practice, this means that the AI works right in the space where people are. The analysis happens inside the building while a situation is still unfolding, so staff can act in the moment.

Privacy is built in from the start: Vision AI doesn’t capture images, and building data can stay on site. Because the sensors don’t capture images, there’s no video data to review, so deployment takes days or weeks instead of months.

The models are also ready before they arrive. Training on simulated scenes prepares VAELLA’s Vision AI and Building AI Operator for busy and unusual situations long before installation.

What the program gives VAELLA

Through the NVIDIA Inception program, our team gets access to NVIDIA developer tools and technical training, preferred pricing on selected NVIDIA hardware and software, and a global network of startups, investors and industry partners.

VAELLA puts it to work where it matters most for customers: sharpening the Vision AI models that run on site, and scaling from single pilots to full stations, terminals and building portfolios.

Let’s connect

If you operate a crowded space and want to understand what your building is doing in real-time, see our use cases or schedule a direct conversation with our team. Talk to us.