January 2022 Summaries

apply(meetup) is a free virtual event scheduled for February 10th from 8:30 am to 1:30 pm PT, bringing together industry thought leaders and practitioners to discuss practical data engineering challenges in building ML for the real world. The event features a packed lineup of speakers, including Erica Greene, Vitaly, Tzvetelina, Erik Bernhardsson, Stefan Krawczyk, Pardis Noorzad, Joe Reis, and others, with topics ranging from model calibration to data engineering workflows. The event is part of the apply() community's efforts to bring together industry professionals and practitioners for discussions on real-world data engineering challenges. Additionally, a pre-announced follow-up event, apply(conf), scheduled for May 18-19th, will introduce new features such as workshops and in-person social gatherings, allowing attendees to finally meet the community in person. Register now at applyconf.com to reserve your seat.

A feature store is a critical component in the development of machine learning (ML) platforms as it enables practitioners to efficiently build production ML systems. It addresses the challenges of managing data pipelines and provides a standardized way to serve features to models in real-time for inference at high scale and low latency. The key considerations when designing a feature store include gathering requirements, understanding the components, and making overall best practices throughout the process. A feature store typically consists of several components including a build, feature registry, data processing engine, orchestration, offline feature store, online feature store, serving infrastructure, access controls, compliance capabilities, SDK, monitoring, canary testing, and hidden challenges. Building a feature store requires careful planning, training, and ongoing maintenance to ensure its success.