I Have a Bone to Pick With Feature Stores

We've all heard the (in)famous Forbes quote about data science ad nauseam, based on a survey of data scientists in 2016: "Data preparation accounts for about 80% of the work of data scientists". The recent advent of the Feature Store is the first meaningful progress toward a different working paradigm for these high-valued employees, buoyed by popularity stemming from success stories at massive tech companies like Spotify, Uber, and Google. However, as the Feature Store has burst onto the scene, it has only magnified the gap in data capabilities and maintenance requirements between the FAANGs of the world and everyone else. In this talk, we'll cover specific examples from thousands of data scientists and data engineers on the technical challenges plaguing the majority of the market. Specifically, we'll cover the three key themes we've seen emerge:

1. Data Scientists lack tooling to work within and take advantage of the Modern Data Stack, built around a centralized cloud data warehouse

2. Transforming data into a feature-ready state is the biggest bottleneck data scientists face to delivering business value

3. Data Scientists do everything in Pandas, but it's up to non-data scientists to refactor everything into production-ready solutions. Additionally, you'll learn about the critical architectural decisions that went into developing the first genuinely cloud-native and modern feature store solution built from the ground up for the Modern Data Stack, including the specific implementation best practices of delivering business value with a feature store solution.