Artificial Intelligence for real-time streaming data
StreamAnalyzer is an enabling platform (iPaaS) that can significantly reduce the cost of development of cloud-native analytics applications for 5G/6G and energy use-cases, amongst other such as for smart mobility. StreamAnalyzer eases the development of analytics applications with a focus on real-time data collection and processing. Users concentrate on building analytics flows while the platform handles such technical complexities as security, privacy, high-availability, scalability, efficient resource scheduling, data visualization, and integration with External Systems, either they act as data providers or data consumers.
Technology
StreamAnalyzer is a general-purpose real-time processing software. It allows receiving events from through multiple messaging protocols. Events are forwarded to a custom network of distributed Spark processors. A processing network is a collection of filters, transformers (AI models are loaded as Spark transformers) and aggregators. Resulting Kafka messages are stored and/or emitted via output APIs. The system is designed for cloud infrastructures supporting horizontal scalability and full multi-tenancy. The platform has been designed to operate on a cloud-native environment. It implements a share-nothing architecture that allows it to scale to a very large number of slices, subscribers, input streams and traffic rate.
High level architecture
We already support integration with 5G Core CNFs and their substrate cloud substrates. Other external systems can inject large and complex volumes of real-time data via their APIs (highlighted with cyan) and we send this data to corresponding Kafka topics.
The outputs (highlighted with gold in the right of the diagram) are sent to external systems and we also provide Graphana visualizations. Our expert software engineers collaborate with the customer to develop the necessary glue code between StreamAnalyzer and the customer’s platform/application/service.
Extensible AI models repository
The AI models database of StreamAnalyzer is customisable and extensible. It currently supports implementations of a wide range of Deep Learning frameworks, including CNN, LSTM, RNN, GRU and their hybrids best fitting specific problems. Moreover, we have implemented an in-house Deep Reinforcement Learning framework (DeepRM) catering for resource provisioning problems in multi-agent Edge telco environments.
StreamAnalyzer employs state-of-the-art open source technologies. Together with superior large-scale resource management in the cloud and a management interface, StreamAnalyzer offers a coherent software stack that enable building of next-generation cloud-native solutions.
Enabler for 5G Core and Edge intelligence
The platform is ideal as the integration layer for either 5GCs or for the 5G Edge. For the latter, it’s built-in support for low footprint messaging protocols renders it an ideal candidate for data collection and processing.
StreamAnalyzer mission: acting as enabler for other verticals
Selected features/modules of StreamAnalyzer have been commercially used in other verticals. Our custom CNN-LSTM hybrid with hyperparameter tuning by our experts has been commercially exploited to forecast power generation of a commercial PV site. Its input includes PV power consumption, on site measurements (incl. temperature, cameras feeds) and satellite images. Its outputs are by ~92% accuracy 15-mins-ahead predictions on generated power.
Strategic direction of the company is to capitalize on the emerging mobility market. We are pursing collaboration with smart mobility stakeholders and investors.