RAHUL ANANTMathematics & Computing
Back to Projects Library
Healthcare AIResearch Prototype & System2024 — 2025

Samarth AI Rehabilitation — Intelligent Healthcare Assistance

Role: AI/ML Systems Lead & Computer Vision Architect

Computer vision-driven physical rehabilitation platform analyzing patient exercise mechanics with real-time feedback and clinical adherence telemetry.

Verified Project Links & Repositories
2 Active Resources

GitHub Repository (Main)

Primary verified codebase, implementation, and commit history

Technical Documentation

Architecture diagrams, API references, and setup guide

Samarth AI Rehabilitation — Intelligent Healthcare Assistance

Technology Stack & Libraries

PythonMediaPipeOpenCVPyTorchFastAPIReactTailwind CSSNumPy
The Challenge / Problem

Context & Objectives

Post-surgical and stroke physical therapy requires continuous monitoring for proper joint flexion angles, but clinical supervision is geographically and economically inaccessible to thousands of patients.

Architectural Solution

Engineered Approach

Engineered a lightweight, edge-computed pose-estimation pipeline running in real-time on standard webcam feeds, calculating Euclidean vectors across critical joint chains and computing biomechanical compliance scores.

System Architecture & Pipeline

Input video stream -> MediaPipe Pose Pipeline -> Biomechanical Angle Calculator (Trigonometric Vector Analysis) -> Anomaly Detection Engine -> Real-time Audio/Visual Feedback Engine -> Clinical Progress Dashboard.

Samarth AI is an intelligent rehabilitation platform designed to democratize high-fidelity physical therapy monitoring. Utilizing on-device computer vision and kinematic modeling, the system tracks 33 skeletal landmarks without requiring expensive specialized hardware, delivering immediate auditory and visual corrective cues to patients recovering from neuromuscular injuries.

Engineered Capabilities & Innovations

33-point real-time skeletal landmark tracking at 30+ FPS on edge hardware
Automated joint angle calculation (flexion/extension/abduction) with ±2° precision
Instant auditory corrective cues for compensation prevention
Longitudinal patient recovery telemetry and compliance reports for physiotherapists

Empirical Results & Benchmarks

Achieved 94.2% exercise form classification accuracy across 5 key rehabilitation movements while maintaining sub-35ms frame-processing latency on standard commodity hardware.

Have Questions About This System?

I am always glad to discuss technical architecture, benchmarks, or potential collaboration.

Discuss This Project