| ---
|
| title: Sentinel Vision Navigator
|
| emoji: 👁️
|
| colorFrom: green
|
| colorTo: blue
|
| sdk: static
|
| pinned: false
|
| license: apache-2.0
|
| short_description: "AI camera guidance for blind users."
|
| ---
|
|
|
| # Sentinel Vision Navigator
|
|
|
| Sentinel Vision Navigator is a web + Android assistive AI prototype for blind and partially sighted people.
|
|
|
| It uses live camera frames, voice commands, Akash-hosted Qwen multimodal inference, and a small vision-navigation RAG layer to provide short, calm, physical guidance such as:
|
|
|
| - Stop. Chair directly ahead, one step away.
|
| - Doorway ahead, slightly left. Take two small steps forward.
|
| - I cannot see enough. Hold still and slowly pan left, then right.
|
|
|
| ## Demo
|
|
|
| - Web app: https://amdvision.qubitpage.com/
|
| - Android APK: https://amdvision.qubitpage.com/downloads/sentinel-vision.apk
|
| - Whitepaper: https://amdvision.qubitpage.com/downloads/sentinel-vision-whitepaper.pdf
|
| - Pitch deck: https://amdvision.qubitpage.com/downloads/sentinel-vision-pitch-deck.pdf
|
| - Model/progress page: https://huggingface.co/lablab-ai-amd-developer-hackathon/SentinelBrain-14B-MoE-v0.1
|
|
|
| ## Hackathon Fit
|
|
|
| - Primary track: Vision & Multimodal AI
|
| - Secondary track: AI Agents & Agentic Workflows
|
| - Supporting track: Fine-tuning / AMD MI300X progress via SentinelBrain model artifacts
|
| - Hugging Face category: Space deployment for public testing
|
|
|
| ## Safety
|
|
|
| This is an assistive prototype, not a medical device and not a replacement for a cane, guide dog, trained mobility support, or human assistance. It is designed to be conservative: if it is uncertain, it should ask the user to stop and pan slowly rather than invent a route.
|
|
|
