Dev Feature Priority Decision

Feature Priority Decision

Decision

The top feature to build next is AI Culling Assistant.

Why AI Culling Assistant comes first

AI Culling Assistant has the strongest end-to-end foundation in the current codebase:

• Batch processing already generates quality scores alongside embeddings and metadata.
• The analyze/index workflow already supports scope selection, progress handling, and writing results back into Lightroom.
• Advanced Search already exposes a quality-based result flow (prettiest / ugliest) and creates Lightroom collections from ranked results.
• Existing metadata fields such as overall_score, composition_score, lighting_score, motiv_score, colors_score, emotion_score, and quality_critique are already stored in the backend.

This means the remaining work is mainly productization:

• turn existing quality signals into a dedicated culling workflow
• define stronger ranking and filtering rules
• create clearer Lightroom collection outputs such as picks, rejects, and shortlist candidates
• improve UX around review and confidence

Why not Duplicate Finder first

Duplicate Finder is promising, but its core grouping implementation is still missing.

• The API route for similarity grouping already exists.
• The service layer already delegates to Chroma grouping.
• The actual grouping function group_and_sort_images(...) is still explicitly not implemented.

That makes it a good second priority, but not the fastest path to a polished user-facing workflow.

Why not Shoot/Event Grouping first

Shoot/Event Grouping has useful input signals available today, but it is less product-ready than culling.

• Capture time, folder names, GPS, and AI metadata are already available.
• There is no dedicated event-grouping service, grouping API, or Lightroom workflow for turning these signals into usable event buckets yet.
• The feature still needs the grouping logic and the user-facing interaction model.

Compared with culling, more of the product shape is still undefined.

Recommended implementation scope

Build AI Culling Assistant first with a narrow first release:

1. Implement burst and near-duplicate grouping first.
2. Use explainable technical and face-aware signals as the initial ranking basis.
3. Add a dedicated command that creates Lightroom collections for top picks and weaker candidates.
4. Allow optional thresholds or target counts per selection/view.
5. Keep the first version deterministic and review-friendly before adding a lightweight aesthetic model.

Priority order

1. AI Culling Assistant
2. Duplicate / Near-Duplicate Finder
3. Shoot / Event Grouping