Fully Baked MLOps

Bake Off

Luke Marsden did what he does best. He wowed us at the meetup last week. One particularly interesting moment was when he tried to present two different ways of containerizing ML models and the pros and cons of each. This ended up in a bit of debauchery as he wanted to stay nonbiased but told us how he really feels about baking.

Check out the slides from the presentation here, follow along with the demo here, and don't forget to have a look at some of the cool open source tools Luke has created namely chassis.ml and boxkite.

Vector Search with Pinecone

I really enjoyed this episode, because Demetrios and I got to do with a good friend of mine, Dave Bergstein at Pinecone! Dave and I worked together at Tesseract Health, and we had a great catchup session about what he's doing now at this exciting new company.

Pinecone is helping pioneer an exciting new category in MLOps: embedding search. Dave explained to us the important utility of this category in industries like e-commerce. We struck a great balance of technical and non-technical subjects.

On the technical side, Dave took us through the engineering challenges inherent to vector search and why an OSS solution or a service like Pinecone can make a difference. On the non-technical side, we got a great story about how Dave's dog is his test case for every Pinecone demo.

I recommend checking out this episode to learn about platform development, vector search, and so much more MLOps. Thanks to Dave for joining us and sharing info about his journey and his work.

Till next time,
Vishnu

We talk to Shalom, an ML engineer at Tribe AI, about quitting his full time job to pursue his startup idea.

How did you know that being a founder was the path for you?
I never wanted to be an employee. I’m so passionate about the work I do. The focus, the long hours – working at a company, I know I’m not going to get the value out of doing that work. The company will. There’s no reward in proportion to your successes. I’d rather experiment and find my own idea.

So you’ve been a founder in search of your idea. What’s been most challenging?
Watching your bank account dwindle. It feels like you either succeed or you get a full time job in a few months. But I kept thinking – how about a third option where you have more time? That’s when I became interested in consulting. However, developing your network and promoting yourself is a full time job I didn't want to do. With Tribe, I don’t have to deal with any of that, which is really nice.

What’s the best part of working with Tribe?
Tribe basically extended my financial runway forever because I have a steady flow of income. It also helped me so my machine learning doesn’t get rusty. My business idea is software, but I didn’t want to spend years not doing ML. I want that part of my repertoire to stay fresh.

What’s your startup idea?
I met another founder and we’re working together on something to automate a legacy industry. There’s so much manual paperwork shuffling out there. All these companies are software companies who don’t know they’re software companies. It’s all really new, but we have investors lined up. It’s pretty exciting.

What comes next?
I’m going to take a break from consulting for a while. That’s one of the coolest things about Tribe. How you can shuttle in or out depending on what else is going on.

Purely OSS ML Platform: Community member Max Hoffman of Dolthub wrote a great blog post about how to build an end-to-end ML platform without using enterprise vendors. Interesting experiment!
Congrats MLFlow!: MLFlow is now a member of the Linux Foundation. Looking forward to their long-running project become even more sustainable.
Why we need feature stores: Community member TiDu Nguyen asked a fundamental question that kicked off a great discussion.
Training on production data before deployment: Community member Evan Peterson asked about prod training cycles and got a great set of practical answers worth reading.
Late night Jamz