Kubeflow Pipelines and Reducing Technical Debt
Members of the MLOps London Meetup gathered again in September to eat, drink, network and learn from each other. In what has become the traditional format, we had two short talks followed by lots of great questions.
Our first presenters were Jonas Mende and Paolo Ambrosio from Sky. The engineering team at Sky had actually been interested in speaking since the inaugural MLOps London Meetup. However, we needed to wait until the open source project had been released and the team was cleared to speak about it.
Thankfully it was well worth the wait! During the talk they covered how and why they built the kfp operator. The kfp operater is an open-source tool that Sky has developed to bridge the gap between Continuous Training and Continuous Delivery on Kubeflow Pipelines. There were some really interesting experiences shared about how Sky abstract the infrastructure and operations away from data scientists.
Following a short break, we had a talk from Lazslo Sragner titled “Clean Architecture: How to structure your ML projects to reduce technical debt”.
Lazslo runs Hypergolic, a boutique consultancy in London specialising in Machine Learning Product Management. He’s also well known in the community for the generous help he offers to people struggling with MLOps issues.
During his talk, he highlighted how important it is to educate data scientists on solid programming fundamentals in order to create a robust ML discipline at your company. He emphasises that cultural change is the hardest part of any transformation and showcases some great techniques for future-proofing your machine learning practice.
Our next Meetup will take place on 29th November – keep an eye on the Meetup Page, details will be out very soon.