Hi, I'm
Pramish

The rise of remote learning has increased the need for effective online exam supervision. This paper presents a web-based automated proctoring solution using a camera and lightweight backend to minimize constant manual monitoring. The system integrates three modules: a YOLOv8s vision model for detecting faces and restricted objects (phones, books), a Dlib facial recognition module for eye and head gaze estimation, and a browser monitor for detecting tab-switching or window manipulation. A rule-based layer analyzes these signals to flag violations, send real-time alerts, and escalate on repeated infractions — with all events logged to a MySQL database for post-evaluation review.

This work explores abstractive summarization of news articles using Facebook's BART (Bidirectional and Auto-Regressive Transformer) model. Unlike extractive approaches that select existing sentences, this system generates concise, semantically coherent summaries by understanding the full context of an article. The model is fine-tuned on news corpora to produce high-quality, human-readable summaries — demonstrating strong performance in retaining key information while drastically reducing article length.