Grounded Research: From Google Brain to MLOps to LLMOps — with Shreya Shankar of UC Berkeley

**Alessio Fanelli** (0:10)
Hey everyone, welcome to the Latent Space Podcast. This is Alessio, partner and CTO in Residence at Decibel Partners. I'm joined by my co-host Swix, writer and editor of L Space Diaries.

**Swyx** (0:20)
Yeah, it's awesome to have another awesome guest, Shreya Shankar, welcome.

**Shreya Shankar** (0:25)
Thanks for having me, I'm super excited.

**Swyx** (0:27)
So I'll enter your formal background, and then you can fill in the blanks. You are a BS, MS, and then PhD in Computer Science at Stanford.

**Shreya Shankar** (0:36)
I'm a PhD at Berkeley.

**Swyx** (0:39)
A PhD at Berkeley, I'm sorry, oops.

**Shreya Shankar** (0:41)
No, it's okay, everything is the bay. Actually, I shouldn't say that, somebody is gonna get mad, but I've lived here for about eight years now.

**Swyx** (0:50)
And then, Internet at Google, Machine Learning Engineer at Viaducts, an OEM manufacturer, an OEM analytics platform, and now you're an EIR, Entrepreneur in Residence at Amplify.

**Shreya Shankar** (1:02)
I think that's on hold a little bit as I'm doing my PhD. It's a very unofficial title, but it sounds fancy on paper when you say it out loud.

**Swyx** (1:10)
Yeah, it is fancy. Well, so that is what people see on your LinkedIn.
What should people know about you that's not on your LinkedIn?

**Shreya Shankar** (1:16)
Yeah, I don't think I updated my LinkedIn since I started the PhD. So I'm doing my PhD in databases. It is not AI machine learning, but I work on data management for building AI and ML powered software. I guess like all of my personal interests. I'm super into going for walks, hiking, love trying coffee in the Bay Area. Recently, I've been getting into cooking a lot.
I feel like I really like pastas, but that's because I love carbs.
I don't know if it's the pasta as much as it's the carb.

**Swyx** (1:55)
Do you ever cook for like large dinners, large groups?

**Shreya Shankar** (1:58)
Yeah, we just hosted about like 25 people a couple weeks ago and I was super ambitious. I was like, I'm going to cook for everyone like a full dinner.
But then kids were coming and I was like, I know they're not going to eat tofu. The other thing with hosting in the Bay Area is there's going to be someone vegan. There's going to be someone gluten-free. There's going to be someone who's keto.

**Swyx** (2:20)
Good luck. You forgot the seed oils.
The seed oil disrespecters.

**Shreya Shankar** (2:25)
I know.
So I was like, oh my god, I don't know how I'm going to do this. The dessert, too, I was like, I don't know how I'm going to make everything. Like a vegan keto nut-free dessert.

**Swyx** (2:35)
Just water.

**Shreya Shankar** (2:36)
It was a fun challenge. We ordered pizza for the children and a lot of people ate the pizza. So I think that's what happens when you try to cook for everyone.

**Swyx** (2:48)
The reason I dug a bit on the cooking is I always find like if you do cook for large groups, it's a little bit like an ops situation.

**Shreya Shankar** (2:54)
Yeah.

**Swyx** (2:55)
Like a lot of engineering. A lot of like trying to figure out like what you need to deliver. And then like what's the pipeline is.

**Shreya Shankar** (3:00)
Oh, for sure. You write that Gantt chart like a day in advance.

**Swyx** (3:04)
Did you actually have a Gantt chart?

**Shreya Shankar** (3:07)
I don't know how people do it.

**Swyx** (3:08)
Did you orchestrate it with Airflow or?

**Shreya Shankar** (3:12)
I orchestrated it myself.

**Swyx** (3:15)
That's awesome.

**Alessio Fanelli** (3:16)
Yeah, we're so excited to have you. And you've been a pretty prolific writer, researcher, and you have a lot of great content out there. I think your website now says, I'm currently learning how to make machine learning work in the real world, which is a challenge that everybody is staining right now from the Microsoft and Google of the world that have rogue AIs flirting with people, querying them to people deploying models to production. Maybe let's run through some of the research you've done, especially on MLOps and how to get these things in production.
The first thing I really liked from one of your papers was the three V's of ML development, which is velocity, validation and versioning.
And one point that you were making is that the development workflow of software engineering is kind of very different from ML because ML is very experiment driven. There's a lot of changes that you need to make. You need to build things very quickly if they're not working. So maybe run us through why you decided as kind of those three V's being some of the core things to think about and some of the other takeaways from the research.