Researcher / Doctoral Researcher
Institute for Informatics and Automation Problems, NAS RA · Yerevan, Armenia
- Designed and validated a real-time 3D human pose estimation pipeline for commodity mobile hardware, sustaining 30 FPS on mid-range devices.
- Conducted a systematic comparative evaluation of inertial versus vision-based motion capture, characterising accuracy, drift, and failure modes of each modality.
- Developed occlusion-aware, part-level supervised contrastive learning for skeleton-based gait recognition.
- Built an intelligent VR workout environment coupling an HMD with an external smartphone as the sensing channel, closing a perception–feedback loop across two devices under tight latency budgets.
- Engineered a guidance layer translating pose-derived kinematic error signals into adaptive, context-aware corrective instructions.
- Full experimental ownership: study design, participant instruction, motion-data collection, signal processing, statistical analysis, and validation against reference measurements.
2020 – present
Founder & Technical Lead — Jamie: AI Fitness Coach
Applied research project · independently funded prototype deployed at scale
- Deployed the research pipeline as a public application (25,000+ downloads), yielding a large-scale, in-the-wild evaluation of mobile pose estimation and feedback quality outside laboratory conditions.
- Implemented real-time on-device computer vision (pose estimation, repetition counting) and 3D virtual-trainer agents in Unity/C#, integrated with a Node.js and Python backend.
2023 – 2024
Included for completeness — large-scale software engineering, team leadership, and delivery of production research systems.
