> Markdown version of [/videos/825-mlops-on-kubernetes-exploring-argo-workflows?t=3](https://www.wearedevelopers.com/videos/825-mlops-on-kubernetes-exploring-argo-workflows?t=3). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# MLOps on Kubernetes: Exploring Argo Workflows

How do you scale machine learning models from research to reproducible production? Discover how Argo Workflows orchestrates complex MLOps pipelines natively on Kubernetes clusters.

- **Speakers:** Hauke Brammer
- **Event:** World Congress 2023
- **Published:** November 10, 2023
- **Duration:** 31:02
- **URL:** https://www.wearedevelopers.com/videos/825-mlops-on-kubernetes-exploring-argo-workflows

## Summary

Transitioning machine learning models from flexible research phases to scalable, reproducible production environments is a persistent hurdle. MLOps bridges this gap, acting as the "younger sibling of DevOps" that streamlines the lifecycle from data ingestion to model serving. To manage the deep complexity of executing sequential and parallel pipeline tasks, organizations are turning to Argo Workflows, a cloud-native engine that allows developers to orchestrate multi-stage processes natively on Kubernetes clusters.

By defining workflows as custom resource definitions, Argo models tasks as directed acyclic graphs where each step dictates a specific container execution. This architecture supports inherently polyglot pipelines where Python scripts, bash commands, and machine learning frameworks execute in isolated pods. Using native artifact handling, Argo automates robust data shuffling—fetching datasets from cloud storage and uploading refined models without verbose boilerplate script code. Additionally, it easily accommodates complex data scenarios involving parallel processing, optional execution, and conditional logic requirements.

A major advantage for batch inference processes lies in Argo's resilient retry strategies and customizable backoff durations, preventing overnight analytical tasks from failing silently due to spotty infrastructure nodes. While smaller proof-of-concept projects or teams lacking Kubernetes experience might prefer simpler automation tools, teams already managing distributed clusters benefit immensely from extending their pipeline operations with Argo. Ultimately, standardizing workflow logic via version-controlled configuration fosters rapid collaboration across data scientists, software engineers, and DevOps teams, guaranteeing consistent software delivery metrics.

**Keywords:** argo workflows, mlops pipeline automation, kubernetes orchestration, machine learning infrastructure, model deployment strategies, batch inference reliability, directed acyclic graphs, cloud-native scheduling, crd resource management, data pipeline artifacts, polyglot workflow execution, automatic pod retry strategies, workflow yaml configuration, docker container execution, multi-cluster deployment

## Chapters

1. **Defining MLOps and its role in production systems** (00:03) — How combining software development, data engineering, and machine learning optimizes model lifecycles.
1. **Solving complex pipeline orchestration with Argo Workflows** (05:32) — Why migrating repetitive Python scripts to a Kubernetes-native engine improves deployment scalability.
1. **Understanding the core components of Argo Workflows** (08:10) — How custom resource definitions, controllers, and stateless architectures manage distributed workflow resources.
1. **Translating simple code into Argo YAML configurations** (10:15) — How to map basic sequential programming functions into containerized Argo step configurations.
1. **Managing machine learning data pipelines with Argo artifacts** (14:40) — Moving training data continuously across cloud storage paths using declarative input and output artifacts.
1. **Structuring model training pipelines with directed acyclic graphs** (18:56) — Defining complex training dependencies and parallel execution limits using automated cyclic graph templates.
1. **Ensuring stable batch inference with automated retry strategies** (23:19) — Configuring custom backoff thresholds and retries to handle unpredictable failures in long-running batch predictions.
1. **Evaluating when to adopt Argo Workflows for data pipelines** (27:20) — Why existing Kubernetes maturity and multi-cluster execution needs determine whether Argo is the appropriate framework.
1. **Comparing Argo Workflows with MLflow and Argo CD** (29:56) — How Argo Workflows integrates with delivery protocols and schedules arbitrary containers unlike Python-only alternatives.

## Related Moments

- [Automating discrete component pipelines using Argo Workflows](https://www.wearedevelopers.com/videos/1582-llmops-driven-fine-tuning-evaluation-and-inference-with-nvidia-nim-nemo-microservices) (from "LLMOps-driven fine-tuning, evaluation, and inference with NVIDIA NIM & NeMo Microservices")
- [Introduction to GitOps and the Argo CD presentation](https://www.wearedevelopers.com/videos/108-get-ready-for-operations-by-pull-requests) (from "Get ready for operations by pull requests")
- [Defining LLMOps and its workflow automation benefits](https://www.wearedevelopers.com/videos/1250-from-traction-to-production-maturing-your-llmops-step-by-step) (from "From Traction to Production: Maturing your LLMOps step by step")
- [Solving application deployment complexities using LLMOps pipelines](https://www.wearedevelopers.com/videos/1582-llmops-driven-fine-tuning-evaluation-and-inference-with-nvidia-nim-nemo-microservices) (from "LLMOps-driven fine-tuning, evaluation, and inference with NVIDIA NIM & NeMo Microservices")
- [Differences between traditional MLOps and GenAIOps](https://www.wearedevelopers.com/videos/1535-from-traction-to-production-maturing-your-genaiops-step-by-step) (from "From Traction to Production: Maturing your GenAIOps step by step")
- [Demonstrating an integrated LLMOps cluster deployment environment](https://www.wearedevelopers.com/videos/1582-llmops-driven-fine-tuning-evaluation-and-inference-with-nvidia-nim-nemo-microservices) (from "LLMOps-driven fine-tuning, evaluation, and inference with NVIDIA NIM & NeMo Microservices")

## Related Articles

- [MLops – Deploying, Maintaining And Evolving Machine Learning Models in Production](https://www.wearedevelopers.com/magazine/115-mlops-deploying-maintaining-and-evolving-machine-learning-models-in-production)
- [MLOps – What’s the deal behind it?](https://www.wearedevelopers.com/magazine/125-mlops-what-s-the-deal-behind-it)
- [MLOps And AI Driven Development](https://www.wearedevelopers.com/magazine/82-mlops-and-ai-driven-development)
- [Building AI Solutions with Rust and Docker](https://www.wearedevelopers.com/magazine/494-building-ai-solutions-with-rust-and-docker)

## Related Jobs

- [Data Scientist](https://www.wearedevelopers.com/jobs/ext/1351648-data-scientist) at **Almedia**
- [Devops Engineer](https://www.wearedevelopers.com/jobs/ext/1940926-devops-engineer) at **Bitpanda**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**
- [Machine Learning Engineer](https://www.wearedevelopers.com/jobs/ext/588393-machine-learning-engineer) at **Twilio**
- [Machine Learning Engineer](https://www.wearedevelopers.com/jobs/ext/1355348-machine-learning-engineer) at **TWILIO**
- [Machine Learning Engineer](https://www.wearedevelopers.com/jobs/ext/1597388-machine-learning-engineer) at **ZEISS Group**