Johannes Otterbach — Unlocking ML for Traditional Companies

Deployment

Johannes Otterbach discusses the challenges of deploying machine learning models in a cloud-agnostic manner. He emphasizes the importance of avoiding vendor lock-in by building a tech stack that can operate across various cloud platforms like AWS, GCP, and Azure. This is achieved through tools like Terraform and Kubernetes, which automate deployment tasks and ensure flexibility 1. Johannes highlights the complexity of managing data distribution across multiple GPUs, introducing tools like Squirrel to simplify data access and storage 2.

We have to build a tool stack that really is cloud agnostic. So we can deploy it like on Prem, we can do it on GCP, AWS, Azure, you name it, whatever it is.

--- Johannes Otterbach

This approach allows for seamless integration and deployment, regardless of the customer's infrastructure preferences.

   

Open Source

Open source tools play a crucial role in standardizing machine learning practices at Merantix Momentum. Johannes Otterbach explains their reliance on PyTorch Lightning for its ability to abstract repetitive tasks in ML training, enhancing code robustness and maintainability 3. He also introduces Squirrel, a tool designed to simplify data distribution in distributed systems, which is set to be open-sourced to benefit the broader community 2.

The idea is to make this open source. Exactly. Cool.

--- Johannes Otterbach

These tools not only streamline internal processes but also contribute to the wider ML ecosystem by providing accessible solutions for common challenges.