Projects

Combining Masked Autoencoder with Neural Cryptography for Encryption

The project focused on developing a secure technique for secret image sharing using a combination of TPM, masked autoencoders, and Shamir’s scheme, enhanced by neural cryptography. This innovative approach aims to mitigate image loss due to noise during the secret-sharing process. The technique is particularly applicable to medical imaging and hidden text transfers.

The primary objective is to facilitate the exchange of confidential information securely, ensuring that no unauthorized parties can access the data. By leveraging Shamir’s scheme, masked autoencoders, and neural cryptography, I created a robust model for transferring secret image shares between two parties. Neural cryptography, a novel form of public key cryptography not reliant on number theory, offers the advantage of reduced computational time and memory usage. It enables the generation of a common secret key between the parties involved.

The goal was to enable the secure exchange of sensitive information over public channels while minimizing computational resource requirements and reducing noise-related image loss.

Employee Management Portal | Website | NIT PATNA —

As part of the project, I eliminated manual work, empowered employees to sign in and manage their own data, and upgraded the system using modern technologies such as NextJS, GatsbyJS, SSO, and Google Drive for data storage. Previously, the university relied on manual processes to upload notices, news, events, and employee details, using outdated technologies and manual data entry in the source code. Currently, the main website serves over 3000 students daily, while the portal supports 350+ employees. My role was pivotal in shaping the website's architecture and ensuring its functionality, with a particular emphasis on scalability. I spearheaded the setup of a comprehensive Continuous Integration/Continuous Deployment (CI/CD) pipeline and a deployment server designed for minimal maintenance, streamlining the development process. I architected and implemented a Server-Side Rendering (SSR) application using NextJS, which resulted in a remarkable loading time of just 0.5 seconds, significantly enhancing the user experience. My responsibilities also included the development of end-to-end (e2e) APIs that featured single sign-on(SSO), which not only fortified the security framework but also integrated the Google Drive API for efficient file uploads. My contributions were critical in delivering a robust and high-performing web application that met the dual needs of employee management and user engagement, reflecting my comprehensive skill set in both frontend and backend development, as well as my proficiency in modern deployment and API integration practices.

Admin Portal -

Fogaze —

AI based chrome extension which tells a person’s health by detecting how many times he blinks his eyes working on his laptop

Helpdesk (11 Apr 2020-19 Apr 2020) —

Web App for HackOn Hackathon responsible for self-evaluation of health by interactive quizzes and other realtime updates

E-Yantra COVID-19 (1 Apr 2020-10 Apr 2020) —

Web App for hackathon organized by e-yantra for getting realtime solutions for the situations created due to pandemic

Robotics Club NITP (Feb 2020-Apr 2020) —

Create complete angular web app of HackSlash website

HackSlash Website (Feb 2020-Apr 2020) —

Designed and developed website of Robotics Club NIT Patna

Sentiment Analyzer (March 2020 – Apr 2020) —

Created complete NLP model to detect human sentiments by their expressed words

Railway Security System (23 Jan 2020-24 Jan 2020) —

Web App for Ministry of Railways made for Hack n Make 1.0 and this is selected in the top 5 solutions for the problem statement provided all over India and got the oppurtunity to get into the SIH Finals

SIKKIM EDUCATION PORTAL (18 Jan 2020-19 Jan 2020) —

Web App for Sikkim Government. made for HackNITP 2.0

Plant Disease Detection (Nov 2019 - Dec 2019) —

Created complete AI model to detect diseases in plants with an overwhelming accuracy of 99%