Computer Vision & Mobile Health AI
YogAsana AI is an intelligent mobile application that provides real-time yoga posture correction, biomechanical alignment tracking, and joint-strain detection. Using edge computer vision, it guides safer independent practice without specialized hardware, external visual sensors, or cloud video streaming.

Value Story
Yoga practitioners often practice independently without expert supervision. Poor form during complex postures such as Trikonasana or Chaturanga can lead to joint misalignment, spinal stress, and acute ligament injuries.
The solution runs real-time skeleton tracking directly on mobile hardware, calculating joint-angle accuracy against anatomical baselines to provide immediate visual and audio alignment cues.
Edge inference on the smartphone camera feed removes the need for expensive motion-capture wearables while keeping visual data on the user's device.
My Contribution
Engineered hybrid pose-estimation pipelines balancing MediaPipe and OpenPose topologies for accurate skeletal keypoint inference.
Built real-time angular analysis algorithms to evaluate joint alignment and movement tolerances during live poses.
Implemented local mobile edge-inference routines in Flutter to eliminate server-side video-streaming latency.
Designed low-latency visual overlays and contextual audio alerts for mid-session posture correction.
Product Capabilities
Detects 33+ anatomical keypoints with continuous tracking designed for partial occlusion and complex body positions.
Compares shoulder, hip, knee, and ankle vectors with target geometric templates to calculate a live posture-alignment score.
Highlights vulnerable joints when knee deflection, lumbar extension, or another tracked angle moves outside configured safety thresholds.
Context-aware audio cues explain which movement to adjust while the user remains focused on the pose.
System Architecture
Camera Stream Layer captures high-frame-rate video through Flutter camera-controller bindings.
Frame Preprocessing and ROI Layer crops, normalizes, and isolates regional bounding boxes for efficient inference.
Edge Keypoint Engine executes lightweight MediaPipe and OpenPose models on local mobile acceleration hardware.
Biomechanical Logic Layer calculates Euclidean vectors, joint-angle offsets, and posture-hold stability scores.
Visual Overlay and Feedback Layer renders skeletal guidance and delivers contextual audio and haptic cues.
Outcomes
Technology
Work together
I can help define the edge computing architecture, optimize neural network inference on mobile devices, and build production-ready applications for real-world interactions.
More projects
Researchers upload manuscripts and receive structured AI feedback before journal, conference, or supervisor submission.
Voice AI for RestaurantsRestaurant teams use AI voice ordering to answer calls, capture multilingual orders, and route kitchen-ready tickets.
Adaptive Learning PlatformPersonalized learning paths, quizzes, progress tracking, and AI tutor agents guide students toward measurable mastery.
Hiring IntelligenceRecruiters manage job posts, screen resumes, run AI interviews, and review candidate reports from one hiring workspace.
Healthcare AISource-grounded clinical knowledge, speech input, and safety-aware summaries come together for clinician support workflows.
Hospitality SaaSConversational booking flows connect guest reservations, real-time availability, and staff table management.