Podcast

ML System Design Playbook: Fail-Fast Design Docs, Modular Architecture & Data Drift Monitoring

S15E1

Open original DataTalks.Club episode

YouTube Spotify Apple Podcasts

machine learning system design

ML System Design Playbook: Fail-Fast Design Docs, Modular Architecture & Data Drift Monitoring

Original Episode

Use these links for the canonical episode and media sources.

Episode Overview

How do you design ML systems that fail fast, scale with modular architecture, and survive data drift in production? In this episode, Valerii Babushkin — Senior Director of Data, Analytics, and AI at BP, Kaggle Competitions Grandmaster, and author of Machine Learning System Design — walks through a practical playbook for ML system design.

People

Use these links to connect the episode to guest notes.

Valerii Babushkin

Chapter Summary

Use these checkpoints to decide whether to open the source transcript.

0:00 - Podcast Introduction
2:06 - Guest Introduction: Valerii Babushkin background
4:00 - Book Announcement: Machine Learning System Design
7:06 - Design Document Purpose: Fail-Fast Principle
8:39 - Blueprint Analogy: Preventing Waste with Early Design
11:09 - Prevalence of Missing Documentation in ML Projects
14:36 - Shared Design Docs: Alignment, Feedback, and Simplicity
19:01 - Design Doc as a Living Artifact: Maintenance Challenges
24:37 - Accountability & Ownership: Assigning Areas of Responsibility
31:59 - Bus Factor & Risk Assessment: People Dependency Mapping
36:50 - Modularity: Chapter-Based Design Docs and Versioning Signals
41:01 - Incentivizing Documentation: Metrics and Performance Reviews
43:53 - ML Design Template Overview: 16-Chapter Book Outline
47:46 - Monitoring: Detecting Data Drift, Concept Drift, and Prediction Drift
51:59 - Fallback Strategies: Redundancy, Simple Baselines, and Serving Reliability
55:13 - Baseline Solutions: Start Simple to Validate Hypotheses Quickly
58:15 - Resources & Tools: Book Discount, Evidently AI, and Templates
1:00:44 - Contact: Connect with Valerii on LinkedIn