CONCEPT Cited by 1 source

ML platform architecture

Definition

An ML platform architecture is the set of infrastructure design decisions that govern how an organization standardizes the full ML lifecycle — from data access and feature engineering through training, evaluation, and deployment — across many teams and workflows at enterprise scale.

The core tension: ML teams need rapid experimentation velocity, but production ML requires governance, reproducibility, and reliability. A well-designed ML platform resolves this by embedding compliance into the execution layer rather than bolting it on as a gate.

Key Design Axes

1. Modularity — Self-contained, versioned building blocks (modules) that compose into pipelines without tight coupling between teams
2. Orchestration — Central scheduling, execution tracking, and workflow composition (nesting, joining, triggering)
3. Governance — Multi-layer access control (user → domain → column-level) embedded in execution, not separate approval workflows
4. Velocity — Hot clusters, deterministic caching, local dev loops, experimental workflows — all designed to reduce iteration latency
5. Integration — Feature stores, experiment trackers, model registries, monitoring, and serving platforms connected via the platform substrate

Seen In

• sources/2026-06-10-atlassian-architecting-scalable-ml-platforms — Atlassian's ML Studio: three-pillar architecture (composable modules, workflow orchestrator, embedded compliance) serving 120k monthly workflow runs across 100+ teams

(Source: sources/2026-06-10-atlassian-architecting-scalable-ml-platforms)