About

We are a consultancy with deep expertise in Machine Learning, Computer Vision and Software Development. We can help you build state of the art solutions tailored to your problem.

For project development we support two modes of operation:

The delivery of our work is customized to your needs; anywhere from prototypes to fully documented and tested solutions. After analyzing your problem we provide you with time-estimated milestones.

We can also provide executive coaching on machine learning topics, for instance on how to get the most out of ChatGPT, and what you can realistically expect from such models.

You can find examples of the projects we’ve worked on here.

How we work

We develop solutions by working from the end goals to the technology and implementation that can deliver it. To build great solutions we need a deep understanding of the problem, constraints and applications. Every solution requires making trade-offs, and it is crucial to know what trade-offs you are making to arrive at informed decisions. For instance, in building a prototype you optimize for development velocity and learning the lay of the land. In building a production system, you optimize for robustness and maintainability. We place strong emphasis on clarifying the problem up front to ensure we are working on the right things in the right way.

Team

Eyrún Eyjólfsdóttir

Eyrún led a team working on Vicarious’ core technology for object detection and pose estimation, bringing it to the efficiency and robustness level needed for production. She led a second team working on 3D reconstruction and synthetic data generation for training machine learning models. During her PhD at Caltech she wrote software for tracking fruit flies, currently used by several research labs around the world, and developed agent-centric models for classifying and predicting fly behavior. Her superpower is to quickly cut the core of the issues and find pragmatic solutions to hard problems. She is a firm proponent of getting the fundamentals right before developing complex models.

Jaldert Rombouts

At BrainCorp, Jaldert developed a patented solution for large scale robotic mapping which remains the core of what is in use today on 35k+ robots across the world. At Vicarious, he led a team responsible for fast and robust production motion planners, and started a simulation team that was key for development velocity and scalability. Jaldert holds a PhD from the Free University of Amsterdam where he worked on the intersection of Machine Learning and Computational Neuroscience. His key contribution was a new learning algorithm for neural networks to be trained with Reinforcement Learning. Jaldert’s major strengths are thoroughly researching the state of the art before tackling a problem and building high quality code that is easy to use and maintain.

Example projects

University of St. Gallen, Switzerland

A course for MBA students on Large Language Models.

This course explained the state of the art as of early 2024 to students without prior knowledge of Machine Learning and advanced mathematics.

Practically, the course made students aware what the strengths and weaknesses of these methods are, and how to get the most out of them. We have extensive interactive material where you can work with various leading models (OpenAI, Mistral), as well as running such models on your own hardware.

We also worked through how a “ask your documents” application works and how to build a basic example, as well as exploring “low code” options to do this.

ELVA Golf, Iceland

We developed a state of the art system (hardware and software) to capture and analyze golf swings at high spatial and temporal resolution. The solution uses multiple hardware synchronized cameras, high quality calibration and state of the art neural networks. We also helped with building out the team to further develop this system.

Janelia Research Campus, USA

Develop state of the art tracking and prediction models for animal behavior.