Building Machine Learning Systems with a Feature Store

Service · Training & workshops · Engineering & development by Hopsworks

Master production ML systems with the FTI pipeline framework and feature stores.

A practical O'Reilly Media guide for data scientists and engineers to build production-ready batch, real-time, and LLM systems using a unified feature store architecture.

Languages
English

Category: MLOps

Format
O'Reilly Media book (Print and Digital)
Level
Professional / Developer
Delivery
Self-paced reading
Publisher
O'Reilly Media
Primary Framework
FTI (Feature, Training, Inference) Pipelines
System Types
Batch, Real-Time, LLM
  • Unified architecture for Batch, Real-Time, and LLM systems
  • Practical guide to Feature, Training, and Inference (FTI) pipelines
  • Authored by Hopsworks CEO Jim Dowling
  • Published by O'Reilly Media

Overview

Building Machine Learning Systems with a Feature Store is a professional guide published by O'Reilly Media for data scientists and ML engineers. Authored by Hopsworks CEO Jim Dowling, the book provides a practical framework for moving beyond experimental notebooks to scalable, production-ready AI systems. It introduces a unified architecture for three distinct types of ML systems: batch, real-time, and LLM-powered applications. The curriculum focuses on the Feature, Training, and Inference (FTI) pipeline framework, teaching readers how to build independent yet connected pipelines using a shared data layer. By following MLOps best practices, the book guides teams through the incremental development and improvement of mission-critical ML infrastructure.

Core Curriculum

  • Unified ML Architecture: The book provides a framework for building batch, real-time, and LLM-powered machine learning systems within a unified architecture.
  • FTI Pipeline Framework: Readers learn to design and manage independent feature, training, and inference pipelines connected by a shared data layer.
  • MLOps Best Practices: The curriculum includes practical guidance on applying MLOps best practices for incremental development and production at scale.
  • Feature Store Integration: The content focuses on introducing ML systems built with feature stores as the central infrastructure.

Real-World Applications

  • Real-Time Recommendations: Readers explore how to build real-time recommendation engines similar to those used by Netflix or TikTok.
  • Churn Prediction: The guide covers building systems to identify at-risk customers before they leave.
  • LLM and RAG Applications: The book explains how to use feature stores to power RAG and create instruction datasets for fine-tuning LLMs.

Deliverables

  • Practical guide to MLOps\nProduction-ready starter projects\nUnified architecture framework

Use cases

  • Data engineers use the FTI framework to build independent feature pipelines that serve both training and real-time inference.
  • Developers use the LLM systems chapters to design RAG pipelines and instruction datasets for fine-tuning models.
  • Architects apply the unified architecture to consolidate batch and real-time ML workflows into a single governed platform.

Specifications

Format
O'Reilly Media book (Print and Digital)
Level
Professional / Developer
Delivery
Self-paced reading
Publisher
O'Reilly Media
Primary Framework
FTI (Feature, Training, Inference) Pipelines
System Types
Batch, Real-Time, LLM
Core Technology
Feature Stores

How it works

  1. Purchase and Access: The book is available for purchase through O'Reilly Media and other major book retailers in both print and digital formats.
  2. Learn the Frameworks: Readers follow the chapters to understand the theoretical framework of the AI Lakehouse and Feature Store architecture.
  3. Implement Starter Projects: The book provides production-ready starter projects that readers can implement using the Hopsworks platform or similar infrastructure.

FAQ

What types of ML systems does the book cover?

The book covers three primary types of machine learning systems: Batch, Real-Time, and Large Language Model (LLM) systems.

What is the FTI pipeline framework?

The book introduces the FTI (Feature, Training, Inference) pipeline architecture, where independent pipelines are connected by a shared feature store and model registry.

Who is the author of the book?

The book is authored by Jim Dowling, the CEO and Co-Founder of Hopsworks.

Added by AntSpace from public sources (hopsworks.ai).

Related

More services by Hopsworks