Whatever Ops 🤷

Joining Forces

For those who don’t know, MLOps World is a leading community of machine learning professionals based in Canada. Since 2017 they’ve been running an impressive North American local events programme plus their annual big blast conference, which is happening this year on June 14th and 15th. (14-17th)

We’ve spoken at their events in the past, and they’ve been guests on our podcasts. As the global MLOps community starts to coalesce; we thought we could light a fire under the process by forging closer links between our respective communities.

MLOps Community is largely Europe-focused, while MLOPs World counts most of its membership in North America. Over the past year or so, we've repeatedly crossed paths with MLOPs World founder Dave Scharbach at events and forums. When that happens 2 or 3 times its starts to feel like fate! After some discussion, we thought, why not try and bring the two memberships closer together?

And now here we are. I should say this is an alliance, not a merger. We’re going to co-promote and try and boost mutual participation in one another’s events and online platforms. We hope to see lots more people from the MLOps World community posting in our Slack timelines and subscribing to the YouTube channel. And we’ll be actively promoting the MLOps World event agenda while they actively promote our content.

Altogether it should mean more sharing of best practices, success stories, and advice on solving thorny practical problems in taking ML models to production.

This year’s event will feature talks from the likes of Matei Zaharia of Databricks and Brian Amadio of Stitch Fix, along with a wide array of workshops, project presentations, and practical advice sessions.

Check out the full program here or click the link below to get tickets.

If You Can't Stand The Heat...

Chris Bergh the CEO of Data Kitchen had many memorable quotes from our meetup on Wednesday. My personal favorite was when we were talking about "don't trust your cloud provider".

I was fascinated by Chris's point about building a factory around the models and the data and transformations. He is more interested in building a system to do innovative work than the specific jargon being thrown around.

"Honestly I don't really care what Ops term you use, call it your favorite Ops."

The system that you work in is an artifact that you need to manage. For cars it's the assembly line, for software it's infrastructure as code.

Working on a shared technically difficult problem there will be some things that are important no matter what industry you are in. Whether its building cars in a factory, using agile or scrum methodology or productionizing ML models you need a few basics. Chris gives us some of his best practices in the conversation. have a listen or watch to see them all.

Speaking to anyone working on data products Chris reminded us all, "You own the result, not just your piece."

Absolute gems laid all throughout this conversation. Check it out on youtube and in podcast land.

MLOps Mastery with Srivatsan Srinivasan

This week, Demetrios and I spoke to Srivatsan Srinivasan, the founder of the popular AI Engineering channel on YouTube and a senior partner at a major technology consultancy. Sri has a true passion for building ML systems and his channel engages in some of the most detailed, thorough, and complete treatments of the MLOps topics we often discuss in the community. It's no surprise then, that his coffee session with us went similarly!

This conversation became pretty technical. We delved into details of how to set up CI/CD, how to set up the right kinds of tests, which cloud tools stand out to us on GCP, AWS, etc., and many other topics. Sri was a wealth of knowledge on all these fronts! Some tidbits:

1. Fail smart, not just fast.
2. Deploy a useless model over waiting to deploy the perfect model
3. Understand that much of DevOps/Kubernetes/CI/CD is YAML proficiency, which can be intimidating, but not impossibly by any means.
4. AutoML has a ton of potential that all of us are eagerly watching out for.
5. Avoid overengineering ML systems by doing things as simply as possible.

I know many of you have been waiting for a more technical approach to some coffee sessions, and I think this is what you want! Let me know your thoughts on some of the practical tips we discussed (especially about CI/CD and testing).

Till next time,
Vishnu

Common Mistakes

This week is a special meetup with Kseniia Melnikova Data Product Owner at Software One.

We will talk about the model lifecycle and development stages and then analyze the main mistakes that tend to crop up at each stage. Kseniia will also provide us with some common solutions to these mistakes.

As a bonus, we will discuss existing tools for experiment management. Kseniia will walk us through some of the ones she has tried and her pros and cons with them.

Bio:
Kseniia is a product owner for Data/AI-based products. Right now, she is working mostly with numeric data analysis, customer insights and product recommendations.

Previously Kseniia worked at Samsung Research with the biometrics team. She has studied computer science in Moscow and management in Seoul South Korea. One of the most interesting directions of her research was Model Lifecycle Management Systems and Reproducibility.

See you on Wednesday aka tomorrow at 5pm UK / 9am California. Click the button below to jump into the event, or subscribe to our public google calendar.