> Markdown version of [/videos/1050-the-road-to-mlops-how-verivox-transitioned-to-aws](https://www.wearedevelopers.com/videos/1050-the-road-to-mlops-how-verivox-transitioned-to-aws). 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).

---

# The Road to MLOps: How Verivox Transitioned to AWS

Struggling with siloed spaghetti code, Verivox pivoted to a scalable AWS-native MLOps architecture. Discover how they slashed model deployment from months to hours and rapidly unlocked generative AI.

- **Speakers:** [Elisabeth Günther](https://www.wearedevelopers.com/@elisabeth-gunther)
- **Event:** World Congress 2024
- **Published:** August 20, 2024
- **Duration:** 25:48
- **URL:** https://www.wearedevelopers.com/videos/1050-the-road-to-mlops-how-verivox-transitioned-to-aws

## Summary

Verivox, a leading German comparison portal, faced a critical operational bottleneck in 2022: their data science team generated valuable models but relied on fragmented, siloed on-premises infrastructure. Deployment took months due to disconnected responsibilities, spaghetti code, and reliance on external teams for API implementations. To break these silos, the team adopted an end-to-end cloud-native strategy, transitioning from isolated Jupyter Notebooks and R scripts to a robust, Python-based MLOps architecture built inherently on AWS.

Following a four-phase MLOps maturity model, the five-person data science team prioritized repeatability, reliability, and scalability by standardizing on managed services. They engineered two distinct reusable architectural blueprints to accommodate different project needs. For standard live inference workflows, they orchestrated automated Amazon SageMaker Pipelines integrated with a serverless API gateway. For complex legacy applications or batch processing workloads, they deployed containerized pipelines using AWS Fargate and AWS Step Functions. To scale these solutions organization-wide, the team minimized operational drift by leveraging the AWS Cloud Development Kit (CDK), defining all infrastructure as code to deploy generic project templates across staging and production accounts alongside custom isolated developer sandboxes.

Transitioning to a true MLOps culture dramatically accelerated delivery, reducing live inference deployment time from several months to just a few hours. The journey demonstrated that investing heavily upfront in clean code principles, Jenkins CI/CD integration, and comprehensive testing—while initially slowing down velocity—pays massive dividends in long-term model maintainability and defect reduction. Additionally, small teams can effectively manage enterprise complexity by committing to fully managed cloud services and actively engaging external solutions architects. Ultimately, this modernized foundation enabled unprecedented agility; when generative AI demand surfaced, Verivox rapidly provisioned a secure, compliant internal LLM playground using Amazon Bedrock in under a week, a feat made possible entirely by their modernized cloud infrastructure.

**Keywords:** mlops transition strategy, aws sagemaker pipelines, machine learning maturity model, aws cdk infrastructure as code, model deployment automation, aws fargate container orchestration, aws step functions, cloud native data science, cross-functional model ownership, jenkins ci/cd integration, live inference architectures, batch processing workflows, python codebase modernization, amazon bedrock llm deployment

## Chapters

1. **Introduction to the Verivox business and services** (00:03) — An overview of the company's online comparison portal and its mission to save users time and money.
1. **The lifecycle and challenges of machine learning projects** (00:55) — The standard phases of building a machine learning model highlight the need for consistent deployment processes.
1. **Understanding the intersection of data science and operations** (02:01) — Integrating multiple disciplines is necessary to productionize machine learning solutions efficiently.
1. **Defining the four phases of machine learning operations maturity** (04:09) — A sequential guide helps teams move from initial proof of concept to scalable multi-project deployment.
1. **Challenges with legacy operational models and siloed codebases** (07:44) — Transitioning away from fragmented notebook deployments requires overcoming significant bottlenecks and missing cross-team responsibilities.
1. **Rebuilding workflows with basic cloud and automation milestones** (10:17) — Shifting from on-premises to the cloud involves standardizing on Python, introducing continuous deployment, and eliminating manual configuration.
1. **Designing an automated machine learning deployment blueprint** (13:40) — Using managed pipelines and custom deployment templates dramatically reduces the time required to push live inference APIs into production.
1. **Managing legacy workflows using containerization and serverless orchestration** (16:47) — Flexible architectures process batch computing jobs and custom models securely using fully managed container services.
1. **Deep dive into infrastructure as code with deployment kits** (18:24) — Defining scalable infrastructure via code templates enables rapid instantiation of isolated environments across multiple projects and stages.
1. **Key learnings and outcomes from the cloud modernization journey** (22:44) — Building strong foundational architectures shrinks deployment timelines from months to hours and unlocks rapid experimentation with emerging technologies.

## Related Moments

- [Defining MLOps and its role in production systems](https://www.wearedevelopers.com/videos/825-mlops-on-kubernetes-exploring-argo-workflows) (from "MLOps on Kubernetes: Exploring Argo Workflows")
- [Real-world application of MLOps architectural patterns at Wayfair](https://www.wearedevelopers.com/videos/369-the-state-of-mlops-machine-learning-in-production-at-enterprise-scale) (from "The state of MLOps - machine learning in production at enterprise scale")
- [Introduction to DevOps for AI and MLOps](https://www.wearedevelopers.com/videos/1222-devops-for-ai-running-llms-in-production-with-kubernetes-and-kubeflow) (from "DevOps for AI: running LLMs in production with Kubernetes and KubeFlow")
- [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")
- [Navigating the MLOps tooling landscape and vendor decisions](https://www.wearedevelopers.com/videos/369-the-state-of-mlops-machine-learning-in-production-at-enterprise-scale) (from "The state of MLOps - machine learning in production at enterprise scale")
- [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")

## 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)
- [Stephan Gillich - Bringing AI Everywhere](https://www.wearedevelopers.com/magazine/489-stephan-gillich-bringing-ai-everywhere)

## Related Jobs

- [Machine Learning Engineer](https://www.wearedevelopers.com/jobs/ext/1597388-machine-learning-engineer) at **ZEISS Group**
- [Staff, Machine Learning Engineer (L4)](https://www.wearedevelopers.com/jobs/ext/1202639-staff-machine-learning-engineer-l4) at **Twilio**
- [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**
- [AI Operations Manager (all genders)](https://www.wearedevelopers.com/jobs/48263-ai-operations-manager-all-genders) at **envelio**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**