Building a safer world with AI-powered geospatial intelligence

Preligens builds cutting-edge software for intelligence analysts to monitor strategic sites. With 50 data scientists at the time of writing, they have one of the largest deep learning-focused teams in Europe. Valohai has helped them minimize time spent on managing infrastructure for model training.

Preligens is at the cutting edge of deep learning

Preligens builds software for analysts in the defense and intelligence space to recognize and track objects and monitor strategic sites. The combination of computer vision and machine learning plays a crucial role in the future of security. It enables constant monitoring of strategic locations across the globe without the immense workforce it has required in the past.

Preligens’ software draws from a multitude of data sources, including satellite imagery (IMINT), signal intelligence (ELINT), and open-source intelligence (OSINT), to provide the best picture of the situation on the ground.

The R&D team at Preligens is at the cutting edge of deep learning, and solving some of the most complex technical challenges comes with the territory. They have one of the largest deep learning-focused teams in Europe, with 50 data scientists at the time of writing, and they are growing their team rapidly.

Preligens has developed a proprietary framework on top of TensorFlow to industrialize R&D and model creation. Their framework streamlines all the steps between experimentation and production. However, Preligens needed additional capabilities to industrialize MLOps, machine deployment, and job orchestration. This is where Valohai comes in.

To support their growth, Preligens chose the Valohai MLOps platform to handle their machine learning infrastructure needs. There are three areas in which Valohai shines for Preligens’ use case:

1. Shared platform
2. Managed infrastructure
3. Integratable architecture

The Valohai MLOps platform functions as the infrastructure layer for all Preligens’ data scientists. Automatic machine orchestration and automatic versioning of metadata and artifacts are the two cornerstones of their setup.

Thousands of experiments, one shared platform

Preligens’ scale simply requires the team to be organized. At any given time, the team will have many simultaneous experiments running, and tracking them becomes paramount. Valohai automatically tracks every experiment run on the platform, storing all the metadata and artifacts. In many other platforms, experiment tracking must be explicitly defined and with a rapidly growing team that may lead to lost work.

Valohai is an everyday tool for our data scientists. When you have an R&D team as large as we have, how to retain all the information you get from experiments becomes a challenge. Valohai solves that for us with its automatic versioning.

Renaud Allioux – Co-Founder & CTO, Preligens

Within the Valohai platform, all the previous work is shared between the data scientists. New team members can onboard more quickly as they too have access to the experiment history and understand how models have evolved. We’ve worked with Preligens to improve our user interface, including experiment tagging, to support their scale.

Managed infrastructure for hybrid cloud

Deep learning with satellite imagery is resource-intensive. Preligens has chosen a hybrid cloud strategy to utilize compute resources that are best suited for each job, and Valohai supports that.

Building a barebones infrastructure layer for our use case would have taken months, and that would just have been the beginning. The challenge is that with a self-managed platform, you need to build and maintain every new feature, while with Valohai, they come included.

Renaud Allioux – Co-Founder & CTO, Preligens

Data scientists don’t have to know about any of the underlying cloud APIs or consoles but rather just select from a dropdown of the available environments when starting the execution. Typically, a hybrid cloud approach would require many more DevOps engineers, but Preligens has mitigated that with Valohai.

Preligens utilizes Azure, AWS, Scaleway, OVHcloud, and their own on-premise machines for computing. In addition to that, specific use cases require the data to stay within the EU. The infrastructure has to support that requirement.

With Valohai, the administrators can assign team members or projects different resources and permissions and flexibly control cost.

MLOps platform – integrated anywhere

The team at Preligens wanted to integrate the MLOps platform with their proprietary framework, which is used to facilitate work and centralize data, parameters, and other resources. Valohai’s API-first approach allows the platform’s full suite of features to be accessed through the API. Any new features we add to the platform are available through the API, CLI, and Web UI.

Preligens can confidently build additional layers that rely on Valohai as the infrastructure layer.

Summary

Building deep learning models is challenging. Building models for mission-critical earth observation use cases at scale is monumentally so. Considering the challenges that their domain sets, Preligens has focused their resources on what matters most – data science. The partnership with Valohai has allowed Preligens to keep their infrastructure team small while scaling their data science team and with it the intelligence they can provide to their customers.

Preligens is showing us what’s possible with computer vision and what’s required to be a trailblazer. Working together has been an amazing experience. Preligens is helping us help them and ultimately, helping us build an MLOps platform that can support even the most demanding industries and use cases.

Eero Laaksonen – Co-Founder & CEO, Valohai

Preligens is pushing the envelope in geospatial intelligence, and we are proud of our part in this European success story.
To learn more about Preligens, visit www.preligens.com