Podcast
Pragmatic MLOps: Build Standardized CI/CD, Model Registries, Monitoring & Org Best Practices
Open original DataTalks.Club episode
Pragmatic MLOps: Build Standardized CI/CD, Model Registries, Monitoring & Org Best Practices
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 build pragmatic, standardized MLOps across teams without chasing every new tool? In this episode, Maria Vechtomova — an MLOps tech lead and manager with roots in econometrics and early work moving from R to Python — tackles MLOps as an organizational challenge, not just a technology problem.
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
- 1:41 - Episode Overview: Pragmatic and Standardized MLOps with Maria Vechtomova
- 3:07 - Background: Early career in data, econometrics, R to Python, and early MLOps
- 5:45 - Early MLOps stacks: Teradata Aster, custom metadata, and orchestration
- 8:03 - Role Overview: MLOps Tech Lead / Manager of Machine Learning Engineering
- 9:45 - Marvelous MLOps: blog, LinkedIn presence, and content cadence
- 11:10 - Defining MLOps: enablement, reproducibility, and teaching data scientists
- 12:42 - Central MLOps team responsibilities: infrastructure, reusable CI/CD, and
- 14:45 - Toollandscape overload: MAD landscape, FOMO, and organizational challenges
- 16:27 - Pragmatic MLOps: leverage existing infra (Kubernetes, Git, CI/CD) not new
- 18:41 - Essential MLOps stack: version control, CI/CD, registries, model registry,
- 20:49 - Model artifacts & registry options: Artifactory, S3, and MLflow alternatives
- 22:23 - MLOps maturity assessment: documentation, reproducibility, code quality,
- 24:01 - Startup priorities: reproducibility, versioning, traceability as first steps
- 27:06 - Team organization: centralized MLOps vs. embedded feature teams and guardrails
- 29:55 - Standardization: cookie-cutter repos, service principals, and Databricks
- 33:24 - Production best practices: move logic from notebooks to packages and CI/CD
- 34:29 - Implementation timeline: technical build vs. organizational buy-in and permissions
- 35:21 - Securing DevOps buy-in: expose pain, deliver standards, and enable internal
- 38:01 - Team composition: small senior ML engineering team building MLOps platform
- 39:29 - Tool-agnostic skills: learn fundamentals and stitch tools together end-to-end
- 42:53 - Roadmap priorities: monitoring standardization, A/B testing, and LLM pilots
- 45:44 - LLM Ops perspective: hype, cost, GPU constraints, and multilingual limits
- 49:42 - Retail use cases: demand forecasting, personalization, and loyalty programs
- 51:24 - Cross-brand model: centralized MLOps support for smaller brands and cooperation
- 54:05 - Learning recommendations: hands-on projects, MLOps Zoomcamp, and pairing
- 56:08 - Skill balance: ML fundamentals plus software engineering and system design
- 57:14 - Data engineering importance: pipeline design, optimization, and data quality