Kineo accelerates AI delivery and cuts orchestration costs with Flyte
Challenge
Kineo needed a flexible orchestration platform for local-to-cloud ML workflows.
Kineo is a Berlin-based AI consultancy delivering bespoke machine-learning solutions across industries—from global sales forecasting to manufacturing quality control. MLOps Engineer Jan Fiedler supports these efforts by providing the infrastructure that enables Kineo’s data scientists to build, test, and deploy ML applications efficiently.
After two years on Kubeflow, Kineo encountered several operational bottlenecks:
- No local execution: All development runs had to be sent to a cloud Kubernetes cluster, increasing cost and slowing iteration.
- Notebook-to-pipeline friction: Jupyter Notebook code required time-consuming rewrites to become Kubeflow pipelines.
- Limited type support: Kubeflow’s type system lacked robust ML-specific types.
- Heavy maintenance burden: Kubeflow required building many components manually and provided limited local testability.
These constraints inhibited developer velocity, created unnecessary cost, and increased MLOps overhead. Kineo needed a platform that supported local development, streamlined Python workflows, and reduced infrastructure complexity.
“If we asked a question of the Flyte community, we could receive the answer in a couple of minutes. So this was huge.”
Jan Fiedler
MLOps Engineer at Kineo
Solution
Flyte delivered lightweight orchestration, local development, and fast iteration.
Kineo evaluated Prefect and Flyte, benchmarking them against Kubeflow across development experience, type support, testability, and Kubernetes integration. Flyte emerged as the strongest fit due to:
- Python-native workflow authoring—no more rewriting notebook code into pipeline code
- Local execution and automated testing, enabling faster experimentation
- Built-in components that replaced many custom Kubeflow modules
- Lightweight operation, reducing AWS infrastructure costs
“Because Flyte is so much more lightweight than Kubeflow, Kineo has cut the cost of AWS base operations by 50%. More important are the cost savings of being able to develop locally.” —Jan Fiedler, MLOps Engineer
Flyte’s composability, strong type system, and local-to-cloud workflow parity allowed Kineo’s data scientists to move faster with less MLOps involvement. The Flyte community provided exceptional support, accelerating deployment and adoption.
reduction in AWS base operations costs
lower workflow development costs (from $160 → $39 per sprint)
faster onboarding, with Kubeflow pipelines migrated in days
Results
Flyte reduced compute and personnel costs while boosting engineering velocity.
Kineo now develops the majority of its pipelines fully locally, only pushing workflows to the cloud in the final stages of a project—drastically reducing infrastructure costs. Local development, caching, and lightweight deployments shrank the cost of a typical two-week data pipeline sprint from $160 to just $39 (a ~70% reduction).
Flyte also reduced personnel overhead. Because workflows are authored in Python with minimal friction, engineers and data scientists can migrate pipelines and adopt Flyte with minimal training.
“They’re adapting to Flyte super, super quickly because we're just dealing with Python. It just enables the engineers to be much much faster.” —Jan Fiedler
Kineo continues to leverage Flyte as its core orchestration engine, enabling rapid ML development, lower operational costs, and streamlined delivery of AI solutions across industries.
