Castalia Conduct

Cross-platform control and data-acquisition

Castalia Conduct™ brings together ROS, node.js and Electron to power Human-Machine Interfaces that can operate seamlessly across platforms. Our framework supports distributed master-slave control architectures out of the box.

With Castalia Conduct™, implementing Graphical User Interfaces for robotic control, sensor data acquisition, and machine-to-machine communication is a breeze. Thanks to our modular architecture and clear-cut API, new functionality can be added with minimal hustle.

Castalia Conduct™ is built on a stack of cherry-picked tools and frameworks; we use ROS along with carefully designed wrappers that enable straightforward interfacing with sensors and actuators. Data acquisition can be handled directly through the middleware for applications demanding real-time communication or using our Node.js-powered server supporting both HTTP and websockets. The Human-Machine Interface leverages Electron so that the user experience stays the same across different Operating Systems and methods of delivery (pure web, native applications).

Castalia Discern

Data-efficient machine learning

Castalia Discern™ is a set of algorithmic library facilitating data-efficient machine learning methodologies. It is geared towards the next generation of AI, i.e AI that will be applicable in contexts where the data points are few and the datasets are skewed. Applications where obtaining definitive ground truths is impossible or prohibitively expensive are common in industrial settings; from high-value asset maintenance and inspection to quality control there are tons of problems where current breakthroughs in AI cannot be leveraged because of lack of sufficient data volumes.

Our approaches fuse well-trusted techniques such as non-parametric Bayesian models with cutting-edge neural network techniques such as Generative Adversarial Neural Networks and Neural Turing Machines. They enable few-shot learning in image classification and time-series regression tasks and the majority of our algorithms is specially designed to be compatible with low-end embedded platforms. In this way they facilitate high-level machine teaching by non-experts.

Castalia Discern™ models are trained using TensorFlow and are delivered across devices leveraging Onnx-compatible formats.

Castalia Explicate

High-dimensional data management and visualisation

Castalia Explicate™ introduces a step change with respect to complex data storage, visualisation and processing. Focused on applications that involve large multi-dimensional data volumes, such as Nondestructive Testing images, our software platform brings clarity and convenience to expert analysts by allowing them to work entirely in the browser – no need for installations, different OS support, update troubleshooting etc. Castalia ExplicateTM leverages the power of WebAssembly to enable complex computational tasks such as image processing to run smoothly within the browser.

Castalia Explicate™ comes with a fully-fledged Node.js backend supporting Digital Imaging and Communication formats, such as DICONDE. It streamlines operations such as noise filtering, frequency-domain analysis and 3D reconstruction.

Both a cloud-hosted and a self-hosted version of Castalia ExplicateTM are available. The latter is provided as docker image enabling the containerised app to be up and running in minutes. Our solution can be extensively parametrised to suit exactly the end-user’s needs.

Want to know more? Feel free to drop us a line.