With covid cases surging across London (and the rest of Europe), our January MLOps meetup was, regrettably, held completely online. What the event lacked in terms of beer, pizza and networking, however, it more than made up for in the quality of the content.
The first speaker was Jan Teichmann, a Principal Data Scientist at The Trainline. I first came across Jan when I discovered his blog on real-time data platforms on Medium. Immediately I knew it was something the MLOps London community would love to hear more about so I reached out and here we are…
Jan talked about some of the unique challenges they have in trying to do real-time predictions on sparse datasets (e.g. for train capacity prediction) at massive scale across millions of users. He highlighted how they use in-cluster operations on their Spark infrastructure and the benefits that brings them. You can watch Jan’s talk here and read some technical deep dives on his blog.
Next up was Matt Squire from Fuzzy Labs. Sporting one of the boldest shirts we’ve seen so far, Matt gave a fascinating talk on open source MLOps.
Fuzzy Labs is an MLOps consultancy who specialize in building platforms with open source tools. They’ve spent a pretty hefty chunk of time analyzing all the different open source tools available and coming up with their definition of what it means to be “truly open source”. Matt walked us through this and highlighted some of the tools he particularly likes working with. You can view the full talk here and find the accompanying github repo here.