Record Evolution GmbH
Arming Data Scientists with a Quality Toolchain

In the age of edge computing, water pumps are intelligent devices controlled by IoT sensors. However, a pump located in a remote area with poor internet connectivity struggles to transfer data to the cloud, or to computer resources with machine learning capabilities. To reduce the task of data transfer, the pump would need to have logic instilled in it, which would also allow the user to implement various functions and deploy software on the remote device.

This task of managing logic on edge devices is the biggest roadblock facing organizations that implement machine learning or IoT-related technology into their processes. Record Evolution, a Germany-based machine learning consultancy, is overcoming that hurdle by laying the pathway for the world of machines and sensors to intersect with modern data science architectures. The other major hurdle organizations encounter is the acquisition of data itself, according to Dr. Marko Petzold, the CEO of Record Evolution—a startup that is represented by 15 data scientists who were previously physicists or mathematicians.

Keeping in mind the pair of roadblocks, Record Evolution has designed two specific cloud data warehousing platforms, namely Reswarm and Repods. While Reswarm allows data scientists to remotely develop and manage code on devices, like the aforementioned water pump, Repods provides the data warehouse infrastructure needed to receive data streams and build analytical data environments. The former also handles the complex task of data harvesting of edge devices.

Repods’ online data warehousing service blends multiple features of data analysis, transformation and visualisation process in a single tool. “Our vision was to create products that are easy to start and adapt, into any kind of data challenge. We’ve done so, with Repods, which makes it easier for smaller companies to invest into a professional data platform,” says Petzold.

The primary purpose of Repods is to ingest streams of data and metadata, in order to create professional pipelines for a historical data store. Petzold believes most organizations “don’t know where to start” with regards to data acquisition, and Repods fixes those challenges. “Our competitors focus on machine learning aspects and expect data ingestion without mapping sources. We study the root source of the data, businesses processes, and presentation of the data. By covering the whole pipeline of that process, we give our clients confidence that they are investing in a robust solution,” he explains.

Repods provides a workbook-style environment where data scientists can train and validate predictive models into the platform while using their favourite tools and advanced machine learning libraries. This capability is best illustrated by Record Evolution’s project with Continental, a globally-renowned automotive supplier. For years, the R&D team at Continental was challenged, to pinpoint the causes of the noises created by their brake systems. In fact, almost every vehicle that pulls over at a traffic light makes a peculiar sound. Upon utilizing machine learning and big data, Record Evolution executed a four-layered Convolution Neural Network along with a few predictive models, which had to be robust enough to distinguish between sounds emanating from radios, ambulances, and the city. For collecting real-time sensor data from vehicles, a prototypical IoT architecture was designed and implemented. The goal of the project, however, was not to just detect the specific noises, but also predict them. Today, the ensemble model sits on an edge device that Continental uses for test vehicles.

From an integration standpoint, Record Evolution’s non-invasive platforms enable seamless integration into an existing workflow and external usage. Going forward, Record Evolution plans to work with production plants that struggle to deploy IoT solutions due to prevailing homogenous machinery. “Most production plants still use machinery from different ages, which have varying capabilities. We have the competence to deploy IoT solutions in such environments,” concludes Petzold.