Podcast
MLOps Architect Guide: Production Model Monitoring, Data Observability & Tooling
Open original DataTalks.Club episode
MLOps Architect Guide: Production Model Monitoring, Data Observability & Tooling
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 keep machine learning models reliable in production — what should you monitor, where do upstream failures originate, and which tooling decisions actually matter? In this episode, Danny Leybzon, MLOps Architect at WhyLabs and computational statistics alum of UCLA, walks through the practical challenges of production model monitoring, data observability, and tooling trade-offs. Drawing on his path from analyst and product roles at Qubole to field engineering at Imply and now advising customers on.
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 - Podcast Introduction
- 1:56 - Guest Overview: Danny Leybzon, MLOps Architect at WhyLabs
- 3:12 - Career Journey: From paralegal ambitions to statistics and machine learning
- 7:26 - Prior Role: Field Engineer / Solutions Engineer experience
- 8:11 - Role Definition: MLOps Architect as technical-business bridge
- 10:32 - Architecture Advising: Tooling trade-offs and landscape navigation
- 12:50 - Role Popularity: Uniqueness of the “MLOps Architect” title
- 13:50 - Startup Reality: Wearing many hats in early-stage companies
- 15:35 - Demonstrating Versatility: Convincing employers you can do it all
- 18:21 - Hiring Story: Cross-functional interview process at WhyLabs
- 22:04 - Career Decision: Choosing startup risk for growth and learning
- 25:04 - Prioritization Strategy: Focusing on production and model monitoring
- 27:35 - Observability Scope: ETL, data pipelines, and upstream root causes
- 28:59 - Customer Profiles: Production-first vs pre-deployment teams
- 30:39 - Market Education: Shift from “why monitor” to “how to monitor”
- 31:50 - Data Profiling Architecture: WhyLogs, profiles, and Apache Druid backend
- 34:25 - Build vs Buy: Guiding customers on tooling and procurement decisions
- 36:47 - Platform Agnostic Integrations: Serving and inference tooling realities
- 38:01 - ONNX Adoption: Interoperability use cases and industry uptake
- 39:10 - Tooling Trends: Cloud-native stacks, heterogeneity, and vendor lock-in
- 41:00 - Research Focus: Fairness, bias, segmentation over explainability
- 43:07 - Productivity Habits: Inbox zero, workspace windows, and task management
- 45:49 - Career Strategy: Exploration vs exploitation and Thompson sampling analogy
- 50:23 - Skillset Advice: Coding, communication, and being an effective Googler
- 55:50 - WhyLogs vs WhyLabs: Open-source profiling vs SaaS observability