Projects

Parkinson's Care VR System
github slides post
C# Unity TensorFlow React Flask SVM CNN
Georgetown University H2AI Hackathon: 1st Place Grand Prize & Patient Safety Award
  • Tackled medical monitoring challenges by architecting a cross-platform VR system integrating four data sources, resulting in objective symptom quantification with 78% accuracy for improved clinical decision-making.
  • Engineered real-time data pipeline connecting Unity VR applications to ML models (SVM, CNN, Logistic Regression), enabling patient assessment that generated 3x more data points with 92% clinical accuracy.
  • Implemented system architecture integrating front-end (React), VR (Unity/C#), and ML models (Python/TensorFlow), creating an end-to-end solution for Parkinson's symptom monitoring and assessment.
WALL-E
github devpost post
Python React GPT-4 API PyQT5 OpenCV C++ Embedded Systems
PatriotHacks 2024: Triple Winner (Patriot Favorite, Most Likely to be a Startup, Best Cyberpunk Theme)
  • Architected a dual-purpose system using modular Python/React application with separate components for file management and waste detection, enabling digital and physical organization in a single solution.
  • Developed LLM integration pipeline using GPT-4 API with custom prompt engineering, creating a file analysis system achieving 95% classification accuracy with natural language search capability.
  • Engineered hardware/software integration connecting OpenCV vision models with C++ microcontroller code, achieving 90% detection accuracy while optimizing for memory constraints on embedded systems.
Peekabot
github devpost post
Python OpenCV AWS IoT Mediapipe C++ Raspberry Pi Distributed Systems
HackOverflow 2024: Best Robot Hack Winner
  • Designed distributed IoT architecture connecting Raspberry Pi (Python/OpenCV) with Arduino microcontrollers (C++) via command queues, creating a robot platform with 30fps processing and 50ms latency.
  • Implemented secure cloud framework using AWS IoT Core with certificate-based authentication and end-to-end encryption, delivering alerts in under 100ms while ensuring data privacy.
  • Engineered modular architecture separating vision, hardware control, and cloud components, enabling parallel development while maintaining system integrity across 3 distributed subsystems.
NaviguideAI
github devpost post
Java Python TensorFlow ResNet Computer Vision PyAudio ESP32
PatriotHacks 2023: Best AI-Powered Hack Winner
  • Engineered computer vision pipeline using CNN-ViT fusion architecture (ResNet-18 with Vision Transformer), creating a depth estimation system with 92% obstacle detection accuracy while reducing costs by 94% ($2,100+ to $129).
  • Developed custom A* pathfinding algorithm with obstacle density penalties and accessibility-focused heuristics, creating optimized routes through complex environments with 127ms end-to-end latency.
  • Implemented audio-haptic feedback system using PyAudio spatial sound synthesis with ESP32-based hardware, delivering directional guidance with 96% navigation success rate in complex environments.
Optimal Path Navigation
github
Python TensorFlow Hugging Face 3D Mapping Depth Estimation A* Algorithm
  • Pioneered depth estimation framework using Python with pre-trained Hugging Face models, generating 500+ environmental maps while reducing implementation costs by 85% (from $2,200 to $325).
  • Implemented image filtering pipeline with custom calibration algorithms in TensorFlow, improving obstacle detection in variable lighting conditions while maintaining real-time processing on standard hardware.
  • Designed extensible pathfinding system using graph-based routing optimized for depth-map navigation, enabling dynamic route planning that adapts to obstacles while preserving computational efficiency.