Back office ops · Production

Flyte: Lyft's ML orchestration platform powers 1M+ pipelines and joins LF AI & Data

The problem

Lyft needed to orchestrate complex ML workflows for its ETA product, requiring management of large historical training datasets, complex output artifacts, backtesting, frequent retraining, and simultaneous multi-model deployment — with the largest bottleneck being infrastructure procurement and management for models that might not work out.

First attempt

The initial v1 of Flyte used AWS Step Functions as its scheduler, which proved too rigid to extend with new features natively, leading the team to build a container-native scheduling engine.

Workflow diagram · grounded in source
1
Historical data training trigger
trigger
“Large amounts of historical data had to be used to train a set of ensemble models.”
2
Multi-model development and deployment
ai_action
“Multiple models were developed, tested, and deployed simultaneously.”
3
Backtesting and hypothesis validation
validation
“Backtesting a model and validating the hypothesis was a requirement.”
4
Artifact output generation
output
“The output artifacts could be as complex as a generated map or a trained model.”
5
Production retraining loop
feedback_loop
“Once in production, retraining the model frequently was required.”
Reported outcome

By mid 2020 Flyte was powering more than 1 million pipelines at Lyft across ETA, Pricing, Mapping, Driver Engagement, Growth, and Map generation teams, and was contributed to the Linux Foundation AI & Data as its 25th hosted project.

Reported metrics
pipelines powered at Lyftmore than 1 million
Platform adoption growthadoption grew exponentially
total cost of running Flytetotal cost of running Flyte at Lyft was low
Reported stack
FlyteAWS Step FunctionsflytekitAWSGCP
Source
https://eng.lyft.com/flyte-joins-lf-ai-data-48c9b4b60eec
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

By mid 2020 Flyte was powering more than 1 million pipelines at Lyft across ETA, Pricing, Mapping, Driver Engagement, Growth, and Map generation teams, and was contributed to the Linux Foundation AI & Data as its 25th…

What tools did this team use?

Flyte, AWS Step Functions, flytekit, AWS, GCP.

What results were reported?

pipelines powered at Lyft: more than 1 million; Platform adoption growth: adoption grew exponentially; total cost of running Flyte: total cost of running Flyte at Lyft was low (source-reported, not independently verified).

What failed first in this deployment?

The initial v1 of Flyte used AWS Step Functions as its scheduler, which proved too rigid to extend with new features natively, leading the team to build a container-native scheduling engine.

How is this back office ops AI workflow structured?

Historical data training trigger → Multi-model development and deployment → Backtesting and hypothesis validation → Artifact output generation → Production retraining loop.