Beyond our commercial projects, our engineers and researchers actively pursue scientific R&D to push technology and science forward. These are areas we are passionate about — and where we bring deep, hands-on expertise that feeds back into everything we deliver.
Real-time and offline signal processing combining traditional DSP methods with neural network and transformer architectures. Our engineers write production code and conduct scientific research to push the boundaries of what's achievable in biomedical, industrial, and IoT signal domains.
Financial/fraud detection and insurance decision making require explainability. We research and build advanced neural networks where stakeholders can understand why decisions were made, meeting regulatory requirements while delivering superior predictive accuracy for risk assessment and fraud detection.
AI agents for data analysis with a focus on answer accuracy, privacy compliance, and seamless integration with existing BI tools. We augment traditional reporting platforms and develop next-generation analytical instruments — built for trust and governed data access.
Open-vocabulary object detection that goes beyond fixed label sets — recognising novel categories in images and video, as well as detecting patterns in signals and audio streams. Our research spans visual, temporal, and spectral domains for flexible real-world deployment.
Neural networks, transformers, and LLMs applied to chemistry — predicting molecular properties, designing new molecules, and accelerating analytical workflows. We serve general chemistry, healthcare, and cosmetology sectors with models that bridge computational chemistry and modern AI.
Scientifically grounded, state-of-the-art tools for analysing news and media streams — understanding social, scientific, technological, and political trends as they emerge. We do R&D, engineering, and commercial solutions. The goal: detect patterns early, build smart summaries with cited sources, and provide analytical instruments powered by publicly available data.
R&D in training specialized LLMs that can run efficiently on local machines for specific domains. We focus on chemistry, insurance, and custom DSL applications where privacy, cost, and offline capability matter. Our research explores efficient fine-tuning, quantization, and domain adaptation techniques that maintain high accuracy while reducing computational requirements.
A unified electricity market intelligence cockpit for European wholesale power markets — ingesting live data from OKTE, ENTSO-E, and HUPX, tracking DAM prices, cross-border flows, and volatility in real time. Multi-model ML forecast ensembles — including ridge regression, seasonal decomposition, kernel methods, and time-series predictive neural networks — deliver price predictions with confidence bands across short and medium horizons. Models are trained on both publicly available market resources and our own continuously growing proprietary time-series dataset collected from live market feeds, enabling the system to improve its forecasting accuracy over time. Built to replace scattered manual monitoring with a single decision-oriented environment for trading, risk, and portfolio management. Active research and continuous improvement.
Research in running complete AI systems including LLMs, multimodal models, and AI agents entirely on embedded systems without internet connectivity. We develop lightweight architectures, sandboxes, and deployment frameworks for autonomous robotics and embedded applications requiring security, privacy, and offline operation.