Modular AI Integration

Sat, 29 Aug 2026 00:00:00 +0000

Modular AI Integration

RawCull’s modular AI work reorganizes application ownership around stable, focused feature boundaries. It does not replace Vision, CLIP, SAM 3, or EfficientSAM with one new machine-learning model. In this documentation, “model” can mean either an inference model installed on disk or an observable application model that owns presentation state and actions. The refactoring is primarily about the second meaning.

The implementation keeps PhotoAIKit as the reusable AI layer while RawCull owns application composition, settings, catalog identity, task lifetime, culling policy, and SwiftUI presentation. Phases 0–6 establish and migrate this architecture. Phase 7 is the next planned step: extracting burst-analysis computation and cache decisions from RawCullViewModel without moving culling commands or review policy out of the application model.

Burst-Analysis on RawCull

Modular AI Integration

Modular AI Integration