🤖 AI Feature Store Workshop

Workshop Overview

This 4-hour hands-on workshop guides you through building a complete feature store implementation using Datorth. You'll learn to create, serve, and monitor ML features at production scale while avoiding common pitfalls like training-serving skew.

What You'll Learn

• Feature engineering patterns — Batch and streaming feature pipelines
• Real-time feature serving — Sub-millisecond online inference
• Point-in-time joins — Correct historical feature retrieval for training
• Feature monitoring — Drift detection and quality tracking
• Feature discovery — Catalog and governance for feature reuse

Workshop Agenda

Module 1: Feature Store Fundamentals (45 min)

Understand the core concepts and architecture of modern feature stores.

• Why feature stores matter for ML operations
• Online vs. offline feature serving
• Feature store architecture with Datorth
• Training-serving skew and how to prevent it

Module 2: Building Feature Pipelines (60 min)

Hands-on lab creating batch and streaming feature pipelines.

• Defining feature schemas and metadata
• Building batch features with Spark SQL
• Creating streaming features with Flink
• Implementing windowed aggregations

Module 3: Feature Serving (60 min)

Deploy features for real-time inference with low latency.

• Configuring online feature stores
• Serving features via REST and gRPC APIs
• Caching strategies for ultra-low latency
• Handling missing features gracefully

Module 4: Training Data Generation (45 min)

Generate correct training datasets with point-in-time feature retrieval.

• Understanding point-in-time correctness
• Building training datasets with time-travel
• Avoiding data leakage in features
• Integration with ML training frameworks

Module 5: Monitoring & Governance (30 min)

Ensure feature quality and enable discovery across teams.

• Feature drift detection and alerting
• Data quality monitoring for features
• Feature catalog and documentation
• Access control and lineage tracking

Lab Environment

Each participant receives access to a fully configured Datorth environment with:

• Pre-configured Kafka clusters for streaming
• Flink and Spark environments for processing
• Redis-backed online feature store
• Sample datasets and starter notebooks
• Jupyter environment for hands-on exercises

Use Cases Covered

Real-time Fraud Detection

Build features for transaction fraud scoring:

• User spending patterns (30-day rolling averages)
• Transaction velocity (last 5 minutes)
• Device and location features
• Merchant risk scores

Personalization Engine

Create features for real-time recommendations:

• User engagement signals (clicks, views, purchases)
• Item popularity and trending scores
• Collaborative filtering embeddings
• Session context features

Prerequisites

• Basic understanding of machine learning concepts
• Familiarity with Python and SQL
• Experience with data engineering (helpful but not required)
• Laptop with modern web browser

Workshop Formats

Live Virtual Workshop

Instructor-led sessions with live Q&A and hands-on support.

• Duration: 4 hours
• Class size: Up to 30 participants
• Schedule: Monthly sessions (see calendar)

Private Workshop

Customized for your team with your use cases and data.

• Duration: 4-8 hours (customizable)
• Delivered on-site or virtually
• Custom labs using your datasets

Self-Paced Lab

Work through the materials at your own pace.

• Access to recorded sessions
• Lab environment for 7 days
• Community support via Slack

Upcoming Sessions

• December 15, 2025 — 10:00 AM EST (Virtual)
• January 12, 2026 — 10:00 AM EST (Virtual)
• January 26, 2026 — 2:00 PM PST (Virtual)