A manufacturer and supplier of industrial machines wants to provide an additional after sales service to customers: Real-Time machine monitoring that can notify operators on the shop floor in case of suspicious sensor readings. Customers can configure rule based monitoring for their shop floor in a custom backend. The analytics engine is also able to produce real time insights based on advanced analytics. As an additional challenge, as more and more customers adopt this service, the DevOps side of the analytics engine has to be migrated from a monolithic environment into a flexible, cloud based setup in order to account for the individual requirements and challenges that each single customer provides. Prototyped different hosting and operations models in AWS EKS and AWS ECS with AWS Fargate.

• Senior / Lead Developer in a team of 6 people
• Lead Research and Conception for Infrastructure Redesign that:
  • Distributes the monolithic analytics engine into a fleet of small, independent AWS native deployments
  • Reduced the end-to-end processing time of a machine monitoring message by up to 80%

For an eCommerce Platform it is crucial to have a detailed picture of customer behaviour on which business decisions can be based. Either in real-time or from the data warehouse. For that a flexible, scalable, and field-testet solution is necessary which can run in the cloud. Additionally, all browser events need a custom enrichment with business information from the backend in order to provide necessary context e.g. for ?Add to Cart?-events. The webtracking pipeline is managed by using modern DevOps principles: Continuous Integration (CI), zero downtime deployments and Infrastructure as Code.

• Integration of snowplow event-pipeline in cloud based shop architecture
• Day to day operations of event-pipeline at ca. 4 million events per day
• Co-Engineering of custom enrichment in the webshop backend (ca. 1000+ lines of code) and handover of ownership to the backend team
• Setup of custom real time event monitoring (< 1s latency) with elasticsearch and kibana
• Setup of custom scheduling and deployment processes for 5 components of the snowplow event-pipeline

To enrich the shopping experience of the customer and to drive additional sales, the eCommerce platform should be able to recommend customers additional products with Artificial Intelligence (A.I.) models. Two orthogonal strategies are employed: Product based similiarity based on neural network embeddings and collaborative filtering based on user behaviour. The model results need to be integrated into the Java backend of the webshop. Additionally, Performance monitoring for the recommendations is needed.

• Product Recommendation Engines: Collaborative Filtering and Item Similarity with Neural Nets
  • Productionize both models based on proof of concepts by Machine Learning Engineerin including data aquisition, training of the model and data output
  • Scheduling and operations of productionized models, including 3 different code bases and more than 5 regularly scheduled jobs
  • Operationalization of 10+ performance metrics over 5 dashboards for stakeholders

In order to enable an eCommerce organization to become a datadriven organization there must be (among other things) a framework present to compare different version of the website against each other. Many members of the organization and departments need to be able to create and conduct experiments without the assistance of a data engineer. Anther important factor for the framework was the usage Bayesian statistics.

• Leading Conception of testing framework including randomization logic, statistical modelling and grapical presentation in the frontend
• Implementation of proof of concept for statistical engine
• Implementation of production code for frontend, backend, statistical engine
• Training of stakeholders from 3 different departments in methodology and statistical background of A/B-testing