Inspiration
Terabytes of 4K camera raw footage stored on NAS often become "dead data" because locating specific 3-second B-rolls requires hours of manual scrubbing. Existing cloud solutions demand costly 50GB+ uploads, compromising privacy. We built ClipVault AI to turn UGREEN AI NAS into an autonomous, edge-native video asset hub using on-device NPU indexing.
What it does
ClipVault AI runs local scene-cut boundary detection, extracts keyframe thumbnails, and generates offline vision embeddings directly on UGREEN AI NAS. Users query via natural language (e.g., "3 close-up b-rolls of pouring coffee"), and the Agent retrieves frame-accurate timestamps to assemble NLE-ready XML timelines for Premiere Pro and Final Cut Pro.
How we built it
- Orchestration: Built strictly following the Uliya Skill Development Standard (
clipvault-agent). - Media Ingestion: Utilizes high-IOPS NVMe storage pools via
file-agentand FFmpeg demuxing. - On-Device Vision: Runs local ONNX Runtime vision models accelerated by Intel Core Ultra NPU cores.
- NLE Export: Generates industry-standard FCPXML and EDL rough-cut timeline projects.
What's next
In Round 2, we will deploy ClipVault onto the assigned UGREEN AI NAS iDX6011 Pro physical environment, benchmark sustained NVMe multi-stream 4K throughput, and validate real-time assistant voice triggers.
Log in or sign up for Devpost to join the conversation.