Problem: Traditional MoCap demands multi-camera optical studios and costly sensor suits.
System: Client-side platform converting monocular webcam feeds into animatable 3D skeletal armatures (.BVH export for Blender/Unity) & 2D puppet rigs.
Contribution & Result: Built real-time BlazePose-to-Three.js kinematic retargeting and mathematical BVH joint serializer—cutting 3D animation turnaround and costs by 80% with zero cloud dependencies.

Problem: Assistive reading tools fail to detect real-time cognitive difficulty and reading fatigue in neurodivergent users.
System: Privacy-first, browser-native assistive platform fusing gaze tracking with adaptive typographic scaffolding and TTS.
Contribution & Result: Engineered the MediaPipe iris tracking pipeline, extracted behavioral gaze features from the ZuCo benchmark, and trained an XGBoost classifier achieving 0.9989 AUC for cognitive difficulty prediction.

Problem: Airline booking platforms demand robust concurrent seating, instant ticketing, and granular fleet administration.
System: High-performance full-stack flight operations platform with live route search, seat matrix allocation, and ticketing workflows.
Contribution & Result: Designed end-to-end REST APIs, client transactional state management, dynamic aircraft cabin matrices, and automated PDF boarding pass generation.

Problem: Fragmented apps for routines, nutrition, and finance compromise user privacy with unencrypted cloud tracking.
System: Local-first Android suite with automated schedule duration splitting, calorie & hydration logging, and double-entry accounting.
Contribution & Result: Architected Clean Architecture MVVM layers, reactive Jetpack Compose UI, and zero-leak offline persistence via encrypted Room SQLite.

Problem: Unimodal sentiment models fail when interpreting subtle sarcasm, voice inflections, or mixed facial expressions.
System: Low-latency GPU inference pipeline fusing acoustic pitch, facial action units, and semantic textual embeddings.
Contribution & Result: Trained CNNs on RAVDESS with Librosa MFCC acoustic extraction on TensorFlow/CUDA, fusing Wav2Vec2, DeepFace, and Transformers for robust multimodal inference.