zeteoh's mission "Elevate Human Potential Through Data Science" relies on 4 technological pillars. In the same way we are pusruing our mission for our customers, we apply the same principles internally. Our technology makes our engineering more scalable. Which in fine allows us to deliver cost effective and cutting edge solutions to our customers.

1️⃣ Data Quality

Thanks to our unique experience in various industries such as retail, healthcare, infrastructure, gaming, insurance and transportation we are able to create efficient data collection strategies. We leverage this knowledge in all our products to reach state-of-the-art performance with limited costs and time.

2️⃣ Data Efficiency

Among recent research breakthroughs in machine learning, self-supervised and semi-supervised learning are already used as a foundation of our technology. It allows our models to be trained with 90% less data than conventional methods with same or better performance.

👉 Self-supervised Learning
We use these methods to make our models understand the structure of the data without the need of labels.

👉 Federated learning
When privacy is critical, this family of methods ensures high-performance models under constraint. The end-user can safely benefit from decentralized training without risks for its own private data.

👉 Model Calibration
Targeted data collection requires well-calibrated models. The final technological block for data efficiency allows us to train models to recognize correctly when a data point is familiar or not.
Combined all together, our data pipelines automatically collect unfamiliar data from the models in production. Our datasets become increasingly meaningful, allowing our system to train stronger and stronger models. All fully automatically, without any human intervention.

3️⃣ Neural Architecture Search

The nature of the data and the environment in which the models are deployed affect deeply the effectiveness of model architectures. We believe that Machine Learning Engineers should not design models directly but supervise systems which are themselves creating models.
In order to achieve this vision, we leverage Neural Architecture Search. Those AI models are responsible for designing optimized architectures for different missions and constraints.

Those methods enable efficiency at scale.

4️⃣ Efficient Algorithms

In order to realize our vision, our AI models should be ubiquitous and sustainable. We strongly believe deploying models on low-power and cheap devices is essential.
TinyML is a category of machine learning models running on microcontrollers. While a common computer vision model takes a few hundreds megabytes, tinyML model sizes are about dozens of kilobytes, 1000 orders of magnitude smaller. We leverage this technology to provide innovative and affordable products.
Edge Computer in general is our de facto solution for deployments.
All our model architectures are designed and optimized by Neural Architecture Search.


Service Icon

zeteoh PLAY


Service Shape
Service Icon

zeteoh RISE


Service Shape
Service Icon



Service Shape