Podcast
Build Scalable, Reliable ML Systems (MLOps): Design Docs, Data Strategy & Edge Constraints
Open original DataTalks.Club episode
Build Scalable, Reliable ML Systems (MLOps): Design Docs, Data Strategy & Edge Constraints
Original Episode
Use these links for the canonical episode and media sources.
- Open the original DataTalks.Club podcast page
- Watch on YouTube
- Listen on Spotify
- Listen on Apple Podcasts
Episode Overview
How do you design machine learning systems that scale, stay reliable in production, and meet tight edge and mobile constraints? In this episode, Arseny Kravchenko — a seasoned ML engineer focused on computer vision, active in ML since 2015 and a former Kaggle Master — walks through practical MLOps patterns for turning models into production systems.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 0:00 - Episode Overview: Building Scalable & Reliable Machine Learning Systems
- 2:34 - Guest Bio & Startup Experience (deep learning, MLOps, Ntropy, AR, Lyft)
- 6:11 - Startups: ML Productionization Trade-offs and Decision Ownership
- 7:54 - Defining Machine Learning System Design: Goals and Constraints
- 10:34 - Edge & Mobile ML Constraints: Latency, FPS, Energy, Core ML
- 14:49 - Managing Unknowns: Known Unknowns, Unknown Unknowns, Early Tests
- 18:49 - Planning Value: Why a Lightweight Design Phase Matters
- 20:21 - Design Document Approach: Problem-First, 50/50 Problem vs Solution
- 22:48 - Problem Framing: Product Scenarios, Realism vs Appeal Trade-offs
- 29:01 - Goals, Non-Goals & Assumptions: Turning Requirements into Metrics
- 31:42 - Solution Blueprint: Baseline, Metrics, Pipeline Components
- 32:37 - Data Strategy: Availability, Processing, Feature Needs, Data Lakes
- 37:15 - System Diagramming: Data Flow, Dependencies, Real-time vs Batch
- 39:42 - Motivation for the Book: Generalizing Experience into Patterns
- 41:45 - Heuristics for Datasets: Intuition, Limits, and Practical Guidance
- 45:10 - Design Doc Examples: Photostock Search & Super Mega Retail Pricing
- 47:09 - Reader Types: Theory-Focused vs Template-Focused Audiences
- 48:27 - Co-author Dynamics: Balancing Corporate & Hands-on Perspectives
- 51:39 - Book Development: Scope Decisions, Publisher Constraints, Reviewer Feedback
- 55:48 - Favorite Chapter: Preliminary Research, Reuse, and External Sources
- 58:28 - Further Learning: System Design Fundamentals & Software Engineering Skills
- 1:00:00 - Book Offer & Giveaway: Discount Code, Twitter Giveaway Winners