My Projects

A showcase of my work and personal projects

Deepfake Detection System

Deepfake Detection System is an AI-powered desktop application that analyzes faces and detects potential deepfakes. It uses a custom CNN built on top of the ResNet50 architecture and presents results through a clean, dark-themed user interface.

Python

TensorFlow

ResNet50

Tkinter

Custom CNN

Portfolio Website

No Demo available for this project

This portfolio website showcases my skills, projects, and coding journey in a clean and modern interface. I built it from scratch with a focus on responsive design, smooth animations, and usability. I continuously update it as I build new projects and learn new technologies.

HTML5

CSS3

JavaScript

Responsive Design

Git & GitHub

Movie & Theatre Booking System

A full-stack web application that allows users to browse movies, select seats, and book tickets seamlessly. It features a robust backend for booking management and a responsive frontend for an optimal user experience.

React.js

HTML/CSS/JS

MongoDB

Node.js

Express.js

ResumeMatch AI

ResumeMatch AI is a semantic resume-to-job-description matcher powered by NLP sentence embeddings. Instead of keyword counting, it uses all-MiniLM-L6-v2 to compute cosine similarity across all JD requirements — surfacing strong matches, partial matches, and skill gaps with actionable improvement suggestions.

Python

Streamlit

sentence-transformers

scikit-learn

pdfplumber

NLP

QueryPilot

QueryPilot is an AI-powered natural-language-to-SQL analytics platform that lets users interact with a PostgreSQL database using plain English. It features automatic schema discovery, read-only SQL validation, self-correcting queries, and dynamic result visualizations — all powered by Google Gemini.

React

TypeScript

Node.js

Express

Gemini API

Supabase

PostgreSQL

CodeSentra - AI Code Detector

A web app that detects whether a Python code snippet was written by a human or generated by an AI model (ChatGPT, Gemini, Copilot, etc.). It extracts 11 stylometric and CodeBERT-based perplexity features, fed into a 3-layer PyTorch MLP trained on ~180 labelled files, and returns a confidence score with a per-feature breakdown explaining which signals drove the verdict.

Python

PyTorch

CodeBERT

FastAPI

scikit-learn

Render