Cases

When the company first considered deep learning as an approach to weather forecasting, they brought us in to run a proof-of-concept project built on Google's GraphCast model. We took full ownership of the pilot, designing and building a data and training pipeline from scratch, guiding architectural choices, and advising on the hardware and infrastructure needed to make it work. This was the company's first real exposure to deep learning driven forecasting, and our role was to determine whether the approach held genuine promise for their context, as well as help them understand the strengths and weaknesses of the direction. The pilot succeeded, and gave the company the confidence to pivot and commit fully to neural weather forecasting as a core direction.

With the proof of concept validated and the company now fully pivoting toward AI-based forecasting, we took on a leading role in building out their development capabilities — with a strong focus on enabling their team to operate and grow independently. We designed a flexible framework that standardizes how models and datasets are defined, and spent significant time transferring the knowledge needed for their developers to work with it confidently and extend it on their own. Alongside this, we guided the team through key decisions on model architecture and training strategy, helping them build the judgment to make these calls themselves going forward. We also advised on hardware and infrastructure, ensuring the system scales efficiently without unnecessary cost. The goal throughout was not just to deliver a working system, but to leave the team meaningfully stronger — equipped with the skills, structure, and understanding to thrive independently in the fast-evolving field of neural weather forecasting.

Built a custom tool that streamlines how SHN's legal team tracks and processes case law. The system automatically ingests new court rulings daily and uses AI to extract key signals—guilt determinations, compensation amounts, offence types—presenting structured insights to legal experts. A human-in-the-loop approach ensures the team stays in control: reviewing, accepting, or rejecting cases while the AI handles the repetitive analysis. This work led to Rechtlezer, our public tool for navigating Dutch case law.

Delivered a training program on evaluating the human rights impact of algorithms using the IAMA framework. In partnership with ECHO Expertisecentrum Diversiteitsbeleid and Open State Foundation, we explored how technology can strengthen or threaten public values—and how organizations can maintain control over what algorithms drive.