Savin Goyal on Metaflow and the state of ML infrastructure.
Subscribe: Apple • Android • Spotify • Stitcher • Google • AntennaPod • RSS.
Savin Goyal is CTO and co-founder of Outerbounds, a startup building infrastructure to help teams streamline how they build machine learning applications. Prior to starting Outerbounds, Savin and team worked at Netflix, where they were instrumental in the creation and release of Metaflow, an open source Python framework that addresses some of the challenges data scientists face around scalability and version control.
Netflix has always sort of stuck to the idea that better quality data will result in better quality models. So that was the focus. But then, of course, you know, Netflix also operates at the bleeding edge of machine learning. So there were many areas where we were using deep learning, but I wouldn’t say that deep learning was the most popular choice.
The machine learning universe is really fast moving. So how can we make sure that we’re not making a bet, that would hinder our progress, two years or four years further down the line. Deep learning is super popular, but tomorrow there could be a new way of doing machine learning. How can we make sure that we are providing that freedom of choice to our data scientists, and not really encumbering on that.
Highlights in the video version:
- Introduction to Savin Goyal
Metaflow, a Python framework
Python, R, machine learning, and deep learning
Metaflow: support entire pipe from feature engineering to modeling
Functionality of Metaflow
Metaflow and AB Testing
UX for Metaflow
Libraries that data scientists love
Large language models and computer vision
Experiment and dependency management
Are data version control and feature stores a part of Metaflow?
Users and contributors of Metaflow
Data wrangling and processing when using Metaflow
Move data between instances
What is the thesis behind starting Outerbounds?
Is your startup based on the open source project or a managed service version of it?
Are you going to start with Metaflow?
Are you targeting users of modern data platforms?
Is Outerbounds going to be cloud native?
Are you considering other users such as analysts, domain experts, or non coders?
Outerbounds platform and UX
What’s the timeline? When is the reveal?
Are the design partners a mix of industries?
What critical positions you are hiring for?
- A video version of this conversation is available on our YouTube channel.
- “What is Graph Intelligence?”
- “Top Places to Work for Data Engineers”
- Azeem Ahmed: “Data and Machine Learning Platforms at Shopify”
- Christopher Nguyen: “What is AI Engineering?”
- Hamel Husain: “Deploying Machine Learning Models Safely and Systematically”
- Travis Addair: “The Future of Machine Learning Lies in Better Abstractions”
- Che Sharma: “Modern Experimentation Platforms”
[Image: Kozloduy Nuclear Power Plant – Control Room of Units 3 and 4 from Wimimedia]