Projects

Diama ("CODE" in Mandarin) is an innovative browser-based IDE designed specifically for novice developers, featuring a lightweight architecture (50% smaller footprint than conventional editors) with essential development tools. Built on Monaco editor core (the same engine powering VS Code), it delivers professional-grade syntax highlighting and IntelliSense while maintaining a simplified interface. Key features include real-time collaborative editing via WebRTC, full filesystem access through the File System Access API, and interactive algorithm visualizations for enhanced learning.

MentorME is a academic and professional networking platform designed to connect students with experienced peers who have navigated similar challenges. It offers a streamlined interface for posting help queries ("Ments"), one-to-one messaging, and credibility through university database integration—solving the problem of unreliable guidance on traditional platforms like LinkedIn or WhatsApp groups.

Mirror is a face recognition-enabled web authentication service designed to provide secure, real-time user verification during web logins. Unlike traditional methods such as OTPs, Mirror enhances both security and convenience by replacing them with biometric authentication. The system ensures that only legitimate users gain access by implementing advanced liveness detection techniques, such as blink-based intent signaling and eye-closure detection. These safeguards prevent spoofing attempts using photos, videos, or masks, and ensure that authentication cannot occur without the user's conscious action—such as blinking—thereby offering an added layer of protection against forced logins.

Carma is an AI-powered video surveillance solution designed to autonomously detect violent activities in real-time CCTV footage. As urban areas deploy thousands of surveillance cameras, the burden of manually monitoring them becomes both impractical and inefficient. Carma addresses this issue by using deep learning models—specifically LRCN (Long-term Recurrent Convolutional Networks) and ConvLSTM—to classify incoming video feeds into fight or no-fight scenarios. Once a video is uploaded to the platform, it is processed by the LRCN model, which integrates CNN for spatial feature extraction and LSTM for temporal sequence learning. The final output is a labeled version of the input video, indicating detected events of interest.