Our Cases
Discover our latest customer cases
Estimating the history and future of vegetation based on satellite imagery
We partner with Agdir Insight AB, a remote-sensing analysis company delivering ML-powered agricultural insights to worldwide customers. Satellite imagery is powerful but clouds, snow, and multi-day revisit cycles can obscure the optical imagery, leading to less available analysis. Together we built a gap-filling engine that reconstructs daily vegetation index maps, creating a consistent historical record. On that foundation, we further developed forecasting models to predict field conditions one month ahead. The result of the advanced AI solution is clearer signals for agronomists and growers, earlier risk detection, and more confident decision making.
Automatically translating 5 MLOC from Delphi to C#
We're partnering with Danfoss Power Solutions on an ongoing modernization to move a 5 million lines of code (MLOC) Delphi codebase to C#. A manual rewrite was estimated at ~308 developer-years, so we're taking a smarter route. Instead of hand-coding for years, we orchestrate a coordinated, multi-agent AI pipeline that plans the work, translates code, validates behavior, and iterates like a disciplined engineering team. The agents compile, test, and debug automatically to catch issues early and keep quality high, then reassemble the project so the C# result preserves the original logic. The outcome is a faster, lower-risk path from legacy to modern .NET—delivering maintainable, production-ready code without the multi-decade wait.
Searching through the archives of one of the world's largest murder cases using an AI powered chat
We partnered with Spår, one of the largest true-crime podcasts in Sweden! Together we built an AI system where a user can chat with the archives of one of the world's largest murder cases, of Sweden's prime minister, Olof Palme. The project existed of two parts, first, an OCR pipeline was developed that extracts all text, names, places, dates, and descriptions and indexes it into a searchable library. Second, an AI chat that can read the library, pull up the most relevant pages, and reason across thousands of records to surface connections and clues. The entire stack is open-source and runs on our own servers, so the data never leaves our infrastructure and no cloud services are required. The result is faster research, a more controlled development process, and a privacy-first way to explore a complex case.
Intelligent traceability of timber using computer vision
We partnered with Tracy of Sweden to deliver end-to-end traceability of timber that curbs illegal logging, tightens quality control, and streamlines operations. Each log end acts as a unique fingerprint, and our computer-vision pipeline re-identifies the same log as it moves from the sawmill station to the sorting line. The system cleans up images with image segmentation and focuses on the end-grain so changing light, dust, and busy backgrounds don't throw it off. Trained on thousands of real-world examples, it achieves ~95% matching accuracy in production with minimal changes to existing workflows. We're now extending the approach to harvester-mounted cameras, pushing traceability all the way back to the forest.