Training and Ensuring Reliability of ML Models at Wolt: The Power of ArgoCD, Flyte, and Argo Workflows

on YouTube ↗25m

" summary

Wolt runs reproducible ML training on Kubernetes using Argo CD + Flyte + Argo Workflows, ties models into MLflow and deploys with Seldon Core, and automates Gatling load tests with assertions so deployments fail CI if P99 or error rates regress. A dynamic Flyte pipeline trains and load‑tests models per city and runs 460 cities in parallel, producing pass/fail summaries that gate model promotion and catch performance regressions before production.

AI-written summary · about

" chapters (11)

1. 0:00Introduction and context
2. 1:08ML use cases at Wolt
3. 3:49Platform needs and goals
4. 5:22The ML platform stack
5. 7:01Argo CD for infrastructure
6. 9:48Flyte for workflows
7. 12:58Load testing with Gatling
8. 15:42Testing reporting and logic
9. 18:32Why this stack works
10. 19:54Practical implementation
11. 22:14Future work and conclusion

" related talks

• 33mScaling Open Source ML: How Wolt Uses K8s To Deliver Great Food to Mill...Stephen Batifol, Ed Shee · KubeCon + CloudNativeCon Europe 2022
• 51mSpeeding up deployment of machine learning modelsAmine Saboni · Voxxed Days Luxembourg 2024
• 25mMLOps with FlyteSamhita Alla · Kubernetes Community Days Chennai 2022
• 26mManaging ML Models @ ScaleSrivathsan Canchi, Tobias Wenzel · OpML '20
• 25mLeveraging Argo Workflows and Argo Events for Cloud Agnostic Enter...Charles Adetiloye, Alex Lerma · ArgoCon 2022
• 28mRun Your Data and Machine Learning Workflows on Kubernetes with FlyteSamhita Alla · Kubernetes Community Days Bengaluru 2023
• 46mDeploying scalable Machine Learning models with KubernetesAlessandro Vozza · Future Tech 2019
• 18mAutomate the boring ML stuff with pipelinesCatherine Nelson, Hannes Hapke · Scale By the Bay 2020