A tire recall incident highlights the gap between a data scientist's local Python script and a scalable, production-ready solution.
The journey began with a demand forecasting use case, leading to the concept of a lab for experimentation and a factory for industrialization.
A standardized process with dev, QA, and prod stages was created, supported by a cloud-agnostic tool stack including Git, Jenkins, and Kubernetes.
The architecture uses Kubernetes and containerization to create reproducible dev, QA, and prod environments with immutable builds and stage-specific configurations.
A command-line interface is used to deploy a new version of a Shiny application to the dev environment and promote an existing build from dev to QA.
The platform provides developers with access to Elastic Stack for log aggregation and Prometheus with Grafana for metrics to monitor application performance.
A web-based frontend offers pre-configured templates for RStudio and Python, abstracting away infrastructure complexity for data scientists.
Several applications are showcased, including real-time tire monitoring for mining trucks, optimizing material mixing, and deploying scrap prediction models to edge devices.
The Data Science Factory is part of a broader ecosystem that includes a telemetry backbone, image recognition pipelines, and predictive maintenance platforms.
44:40Hauke Brammer
45:45Simon Stiebellehner
30:49Marin Niehues
22:41Christoph Fassbach & Daniel Rohr
41:06Lars Gentsch
34:00Álvaro Martín Lozano
30:33Falk Langer & Lukas Stahlbock
45:58Andreas Christian
.gif?w=240&auto=compress,format)
Jobs that call for the skills explored in this talk.
SYSKRON GmbH
Regensburg, Germany
evoila Frankfurt GmbH
Mainz, Germany
