SilverGait
An AI-powered physiotherapy companion that turns any smartphone into a mobility assessment tool for Singapore's ageing population. Pure computer vision — no wearables, no sensors, no clinic visits.
AI Chat
Video Assessment
SPPB Scoring
Caregiver View
The Problem
- 1 in 3 Singaporeans aged 65+ fall each year, often leading to hospitalisation and loss of independence
- Frailty screening requires clinic visits — many seniors skip them due to mobility barriers
- Physiotherapy is expensive and inaccessible for the 900,000+ residents aged 65+
- Caregivers lack real-time visibility into their loved one's mobility decline
Our Solution
- Phone-camera SPPB assessment using pose estimation — no wearables, no clinic visits
- AI-personalised exercise plans adapted to frailty tier and specific mobility deficits
- Multilingual chat companion (English, Mandarin, Malay, Tamil) with Singlish-aware voice
- Caregiver dashboard with automated safety alerts for falls and emergencies
What It Does
Video Mobility Check
SPPB assessment via phone camera with real-time pose detection and biomechanical metric extraction
Smart Exercises
Tier-personalised exercise plans with live form feedback and progressive difficulty
AI Wellness Chat
Gemini-powered companion with RAG-grounded health education and function calling
Safety Guardrails
Multilingual emergency detection, auto-escalation to caregivers, deterministic scoring
Progress Tracking
SPPB score history, exercise streaks, trend analysis, and weekly reports
Caregiver Dashboard
Clinical view with frailty scores, alerts, care plans, and shareable reports
Technical Documentation
Dive deeper into how SilverGait works — from the LangGraph agent pipelines to the biomechanical scoring algorithms.
Architecture & API
Two-graph LangGraph system, database schema, RAG pipeline, patient safety guardrails, voice cloning for familiar voices, and API endpoints.
Agent Pipelines
Visual flow diagrams of the Assessment Graph (deterministic scoring) and Chat Graph (Gemini function calling to 4 management sub-agents).
CV Pipeline & SPPB
Why keypoint-based pose estimation over full video classification. MoveNet biomechanical metrics, SPPB clinical cutoffs (Guralnik 1994), and hybrid AI scoring.
Clinical Evidence
Peer-reviewed research grounding every design decision — SPPB validity, smartphone gait analysis, Otago exercise programme, CBT-I for sleep, fall prevention.
GitHub Repository
Full source code, setup instructions, and deployment guide. React + FastAPI + LangGraph + Gemini.