Feel Good MLOps Stories

Happenstance

One of my favorite things in the world is hearing about the connections that are made through the community. Connections can take shape in so many ways. Finding someone to mentor you, creating a new open source project, or as this story goes, finding a new place to work!

This is not the first time I've heard about job positions being filled through the community, in September last year Asli Sabanci got a job at Algorithmia through a posting in the jobs channel. This story, however, feels like it has destiny written all over it.

About a month ago, Sam Bean posted in the #jobs channel. He was looking to fill a few MLE positions at the fintech he worked at. End of story. Or not.

Ray Buhr who worked at PayPal at the time saw the ad and thought hm that could be nice. He didn't directly apply though. And this is where I feel the power of the community really shines through.

Skylar Payne asked one of the most controversial questions MLOps history; are notebook ok to use in production? Not only did this set off a storm of replies, guess who wound up sharing opinions in the 60 message long thread?

That's right, Sam and Ray connected over productionizing jupyter notebooks. It's like a match made in MLOps heaven. From there they started chatting, one thing led to another, and boom! Look who's got a new job!

Congrats to Ray, I'm sure PayPal is gonna miss you. Congrats to Sam, I'm sure you are pretty stoked to fill that role. Let's hope you all are able to stay together longer than the average couple during the pandemic.

Celebrate!

Before the meetup, I told Vishnu and David we should wear our best outfits. apparently, Vishnu took that to mean wear a nice button-up shirt as for David and me?

We got dressed up like it was a costume party.

As for the information shared? Well, there was a bit of that too. What Vishnu lacked in threads he made up for in thoughtful reflections from the past year.

Some highlights:

Thanks to a great audience question, we got to reflect on being an effective translator across all the teams that ML often touches, whether business or engineering. It's key as an ML professional to advocate for your methods and clarify to each person the role that they play in the development (and benefits) of the ML product.

We also discussed the value of starting small and iterating. MLOps is not about what's fancy. It's not about the next shiny tool or framework. It's about what works and what gets the job done. This is particularly true now, given all the talk in the community about feeling overwhelmed by the number of tools out there.

Finally, David and Vishnu had a true mindmeld at one point. After an audience question, both quote Don Knuth's famous adage: "Overoptimization is the root of all evil". MLOps practitioners definitely vibe with this!

Let us know what you thought about the chat, how was this session for you?Check out the video here and podcast here for all the shenanigans.

One Word - Jira

Many, if not all of you, will have had experience with the famous Atlassian toolkit: JIRA, Confluence, etc. We had the pleasure of talking to the man baking machine learning into all those powerful products, Geoff Sims! Geoff is a principal data scientist who has spent the past 5 years standing up the machine learning efforts at Atlassian.

My favorite parts of the conversation focused on the organizational challenges that Atlassian had as a company in trying to adopt ML in their product, as well as how the company uses the same ML products it develops.

On the first point, Geoff took us through the trickiness of joining as a data scientist, but transitioning over the machine learning engineering and working more closely with the software engineering group. Geoff shared with us how this occurred particularly as he worked on the feature store problem at Atlassian, where they adopted Tecton. I highly recommend listening to this podcast for anyone who has a background as a pure data scientist and wants to be more involved with MLOps.

To the second point, there's a real elegance to being able to test the algorithms one develops as a user. In this case, Atlassian as a company uses JIRA and Confluence; thus, whenever new ML features are released, Geoff benefits from learning how internal users react to the changes and can debug and iterate on models more effectively. Quite a flywheel!

As one of the world's most visible software companies, Atlassian's vast data and deep product suite poses an interesting MLOps challenge, and we're grateful to Geoff for taking us behind the curtain.

Check out the video, and podcast.

Till next time,
Vishnu

There's A Storm Brewin'

The way Data Science is being done is changing.

Notebook sharing and collaboration is messy. There is minimal visibility or QA into the model deployment process. The industry is starting to take notice.

This week a different Vishnu will talk about building an ops platform that deploys hundreds of models at-scale every month. A platform that supports typical features of MLOps (CI/CD, Separated QA, Dev and PROD environment, experiments tracking, Isolated retraining, model monitoring in real-time, Automatic Retraining with live data) and ensures quality and observability without compromising the collaborative nature of data science.

You can expect us to touch on these points:

Why is MLOps necessary for model building at scale?
What are various cloud based models for MLOps?
Where can ops help in various points in the ML pipeline Data Prep, Feature Engineering, Model building, Training, Retraining, Evaluation and inference?

Bio: With 10 years in building production-grade data-first software at BBM & HP Labs, Vishnu started building Emagin's AI platform about three years ago. The goal? Optimizing operations for the water industry. At Innovyze post-acquisition, they are part of the organization building world-leading water infrastructure data analytics product.

+ As always see you at 5pm GMT / 9am PST tomorrow, Wednesday by clicking the link below. I'll bring enough cake for all of us.