Deep Learning for quality control
During the micro-chip manufacturing process, a number of defects can occur, such as short-circuits, nicks and dust but not all of them cause malfunctioning. We trained a deep neural network capable of differentiating acceptable from non-acceptable defects based on high-resolution images of produced wafers.
Behavioral modelling for customers
Sensor monitoring in agriculture
Successful cultivation of crops depends on a number of variables. In this particular case, we've built a predictive model based on sensor data providing readings on soil humidity, levels of phosphorus, nitrogen and other elements. Combining this data with ground geography and geometry, weather forecast and details of the crop variety enabled the optimization of production yields.
Time-series analysis for predictive maintenance
Customer behavior showed to be highly predictable given basic social-demographic data and a small sample of previous behavior. We have developed solutions for customer segmentation, product recommenders and feedback processing. Such models enable more personalized communication, tailored offers and overall life-cycle optimization
Failure of large diesel engines is highly preventable given proper maintenance. Using sensor data, engine profiling and engine oil analysis data, maintenance can be optimized so that it prolongs engine life while minimizing down-time for maintenance
Ontology based expert system
In pharmaceutical drug-disease research, the most time-consuming step is the review of the existing literature. Using large volumes of publicly available scientific publications, we have developed an expert system that works with researchers, directing them toward relevant context and related work.