Inspiration
Event and wedding photographers store thousands of high-resolution RAW frames on local NAS storage after every shoot. Manually reviewing every single frame under tight client delivery deadlines creates severe post-production bottlenecks. Existing cloud culling services introduce high subscription costs, privacy concerns, and slow bandwidth uploads.
What it does
Shoot-Triage Assistant is an automated AI Skill that runs 100% locally on a UGREEN AI NAS. It scans authorized photo shoot folders, evaluates image quality, detects duplicates, and compares burst sequences against an optional user style profile (TASTE.md). It generates transparent, plain-language review reports (REVIEW.md) and per-image sidecar files (.sidecar.json) explaining why top picks won—without uploading any photo off the NAS or deleting any user file.
How we built it
Built according to the Uliya Skill Development Standard v0.7.3, Shoot-Triage Assistant orchestrates official UGOS system agents:
- file-agent: Handles local file discovery, directory structure creation, and confirmation-based file moves.
- photo-agent: Performs local metadata extraction, sharpness/blur scoring, visual feature embedding, and scene/burst comparisons.
- knowledge-agent: Reads user style profiles (
TASTE.md) and outputs structured review logs (REVIEW.md) and sidecar metadata files (.sidecar.json).
Challenges we ran into
Designing an intelligent culling tool that remains strictly non-destructive was critical. We established hard safety guardrails ensuring that scoring is purely advisory and original file contents are never overwritten or deleted. Additionally, structuring "Why A beat B" burst comparison notes required fine-tuning plain-language logging so that every decision is transparent and human-explainable.
Accomplishments that we're proud of
- Taste-Aware Triage: Created a persistent profile mechanism (
TASTE.md) that lets photographers train the Skill on their personal aesthetic style. - Explainable AI: Delivered plain-language reasoning logs for every classified image instead of black-box numeric scores.
- 100% Rule Compliance: Successfully designed a zero-bandwidth, completely private, non-destructive NAS workflow meeting all Round 1 Uliya standards.
What's next for Shoot-Triage Assistant
Testing batch throughput and style-matching thresholds on physical UGREEN AI NAS hardware during Round 2 finalist deployment, followed by refining interactive TASTE.md profile generation based on past client gallery selections.
Log in or sign up for Devpost to join the conversation.