Defi.

Projects

FallGuard — Video-Based Fall Detection System

End-to-end fall detection system with real-time inference, bounding boxes, and an interactive Streamlit demo.

Nov 2025 – Dec 2025
Python TensorFlow (tf.keras) OpenCV Streamlit
  • Developed an end-to-end computer vision system for fall detection using a custom-trained deep learning model built with TensorFlow (tf.keras), performing simultaneous bounding-box localization and posture classification (fall vs. stand). Designed and trained the model on a 7,000-image annotated dataset, implementing custom data preprocessing, coordinate normalization, and loss balancing to achieve ~99% classification accuracy and reliable real-time inference.
  • Built a production-ready inference pipeline supporting both video streams and single-image inputs, enabling frame-by-frame detection with adjustable confidence thresholds. Deployed an interactive Streamlit web application allowing users to upload videos or capture images via webcam, visualize predictions with bounding boxes, and tune precision–recall tradeoffs live. Optimized the system for local GPU acceleration on Apple Silicon and delivered a clean, demo-ready application suitable for real-world monitoring and portfolio presentation.

HuskerFresh — Campus Food Discovery & Ordering Platform (CornHacks 2025)

Full-stack platform for discovering campus-adjacent food options and submitting pickup requests.

Nov 2025 – Nov 2025
Python 3.9+ Django 5.2 Django REST PostgreSQL (Neon) HTML/CSS/JS
  • Built a full-stack web application during CornHacks 2025 to help University of Nebraska–Lincoln students discover campus-adjacent food options and submit streamlined pickup requests through a centralized platform.
  • Designed and implemented a Django-based backend exposing RESTful APIs to manage users and food requests, with PostgreSQL schemas structured for reliability, fast queries, and future scalability.
  • Developed a responsive frontend using Django templates and custom CSS/JavaScript, delivering a distinctive, user-friendly interface with Space Grotesk typography and a cohesive banana-themed visual identity.
  • Implemented user authentication and session handling to support personalized interactions, request tracking, and controlled access to platform features.
  • Delivered a polished, demo-ready product within a single hackathon sprint, emphasizing clean architecture, clear data separation, and extensibility suitable for real campus deployment and portfolio presentation.

Invoice Management System

Java-based system for managing construction invoices with a structured SQL backend.

May 2025 – Sept 2025
Java SQL JDBC
  • Built a Java-based system to manage and track invoices for construction projects, reducing manual processing by automating database workflows.
  • Designed a relational database schema to store invoices, vendors, and payments while preventing duplicate or inconsistent records.
  • Implemented JDBC-based create, update, and retrieval operations with transaction handling to ensure data accuracy.
  • Implemented input validation and structured exception handling to prevent invalid financial records and improve system robustness under erroneous inputs.