Kieran Lavelle

The Predictive Maintenance Hub (PMH) is one of many products that live in the Enerlytics ecosystem developed by the Uniper Technologies Group. The product aims to help deliver insights and warnings to power plant operators worldwide about issues that are ongoing, have already occurred, or are forecasted to occur within their plant. The aim of this is to save the plant operators money by predicting these events so that they can perform maintenance before a large amount of damage is done to the plant.

I was responsible for technologically leading the development of the product and managing the 8-person team who worked on the ongoing development of the product. I delivered it as a service, and the PMH is served through the Enerlytics platform. The PMH was built on a dockerized, distributed microservice infrastructure connected to a central data lake. The back-end REST API was developed with Python while communicating with a NoSQL database within the data lake.

The aim of the project was to reduce the amount of storage space being
utilized to store documents contained within a data center. These documents could not be archived or significantly compressed due to the nature of the documents.

In order to solve this problem, I developed and trained a neural network with a CNN LSTM architecture from hand-labeled data. To classify images into three potential classes: black and white, greyscale, and color. The data was labeled by a hand labeling tool that I created which showed the document/image to the user in black and white, greyscale, and color. The user would then select the lowest level of color without a loss of information on the document.

This allowed us to reduce the storage space in the data center dramatically without any loss of quality or information.

The Advertising Reporting Service was a microservice that was designed as a GraphQL API written in Python. The service received requests from customers about the advert reporting data they wanted from third-party advertising platforms. The service would then dynamically build a query and submit this to the third-party APIs and format the results once returned.

This allowed the company to have one highly generic and extensible API to manage all of the off-platform reporting for their customers independent of the advertising platform(s) they used. The service was designed so that it could be consumed from other internal services within the platform and is now managing thousands of requests per hour.