ULEBI

Wardni is a real-time clinical night-shift triage workstation built on Node.js 18+ that combines Meta SAM 3.1 (spatial tracking) with Meta Muse Spark 1.3 (clinical urgency ranking: RED, YELLOW, routine). AI handles tracking and alerting, while nurses decide patient care.

5-Module Architecture (wardni-fullstack-v3.1)

1. Vision (vision.html): Tracks .mp4/.mov clips, webcams, or Room IR streams using Class + Visual Description cards, Spatial <0f> clicks/boxes, Exemplar crops, and Patient Reference Photos. Separates patients from staff via an Origin-Locked Bed Zone (>=40% start overlap), negative exclusions, and bedside staff suppression, plus a Privacy Wireframe mode.
2. Voice (voice.html): Transcribes live bedside speech (en-US, fr-FR, ar-DZ) or typed input and ranks voice urgency alone via Spark.
3. Task Buttons (tasks.html): Six-button bedside pad (Emergency Help, Bathroom, Water, Drink, Eat, Pain / Comfort). Emergency Help triggers immediate hardware-first RED dispatch even if AI is offline.
4. Master Dashboard (index.html): Merges Vision, Voice, and Task outputs across all rooms (RED → YELLOW → routine) with a 1-click Spark shift summary.
5. Nurse Escalation (server.js): Invites on-duty nurses with a 15s–120s step timer, auto-rotates on timeout, and fires a continuous siren + global RED banner if unanswered.

Security & Setup Requirements

Hardened Backend: Zero-dependency server.js keeps MODEL_API_KEY server-side, locks Meta URLs, gates /api/* behind WARDNI_PASSCODE + login throttling, blocks SSRF/XSS, saves state/alarms across restarts (ward-state.json), and requires HTTP 200 handshakes for a green LIVE status.
What You Need: Node.js 18+, Chrome/Edge 126+, a funded MODEL_API_KEY from dev.meta.ai, .mp4 test clips, a mic, and an HTTPS Node host for your domain ahead of the Nov 5 (03:00 WAT) deadline.