Featured Project
HyperSharp
Privacy-preserving AI layer designed for enterprise systems requiring secure and scalable intelligence. It is built for enterprise-grade security and compliance environments.
Visit hypersharp.aiComtrade AI is an applied AI research lab focused on solving technically hard problems that require new intellectual property, new algorithms, and original scientific approaches.
We design and develop novel models, new algorithmic frameworks, and research-backed systems that generate proprietary IP. Our focus area is privacy-preserving AI.
Comtrade AI operates with a research-first philosophy,
combining industrial engineering with academic rigor.
Comtrade AI is shaped by a research-first culture, maintained through strong, ongoing ties to the academic research community and continuous exposure to current scientific work in the field. This keeps our thinking close to emerging methods and rigorous evaluation practices, and ensures that what we build is grounded in evidence rather than hype.
In particular, Dr. Martin Molan is an active researcher at the University of Bologna, which helps keep Comtrade AI connected to contemporary research discourse and standards. This scientific work is carried out in an academic context as part of independent research activity by our researchers, and is referenced here to demonstrate rigor and capability rather than to imply organizational ownership.
In practical terms, it strengthens our ability to tackle problems where off-the-shelf approaches fail, and to translate high-quality research into defensible, production-grade IP.
Privacy-preserving AI layer designed for enterprise systems requiring secure and scalable intelligence. It is built for enterprise-grade security and compliance environments.
Visit hypersharp.aiPredictive AI for Satellite Cryocoolers
In collaboration with the European Space Agency and led by Dr. Gregor Molan at Comtrade Research, the RASCOSA project (RotAry Stirling CryOcoolers for Space Applications) tackles one of the most expensive problems in satellite manufacturing: lifetime qualification of cryocoolers, which traditionally requires destructive testing over thousands of operational hours.
Comtrade AI contributed a novel machine learning pipeline for non-destructive lifetime prediction from standard functional-test telemetry. The approach introduces a family of foundation models — capacity-scaled encoders pretrained on unlabeled multivariate time series — together with a Dimension-Aware Neural Architecture Search (da-NAS) strategy tailored to the small-data, high-imbalance regime characteristic of space manufacturing. The same learned representations support both supervised lifetime classification and unsupervised anomaly detection, enabling early-stage quality assessment without sacrificing high-value components.
The work was carried out under ESA GSTP Activity No. 1000039802, with computational resources provided by EuroHPC on the Leonardo supercomputer at CINECA.
Preprint coming soon.
Research-first culture and peer-reviewed work
Original algorithms and proprietary architectures
AI challenges unsolved by off-the-shelf tools
Interested in collaboration or research partnerships?
