Hi, I'mAyushman Singh
Passionate about building beautiful, performant web experiences.
My Education
The academic foundations that built my technical thinking.
Bachelor of Technology
Computer Science & Engineering
Sardar Vallabhbhai National Institute of Technology (SVNIT), Surat
Currently pursuing B.Tech in Computer Science & Engineering. Focused on Data Structures, Algorithms, Web Development, and Machine Learning.
Key Highlights
- Winner - Web Wonders 2025
- Runner up - Enfinity Hackathon
- Finalist - ACM Summer Challenge
- Runner up - Google Winter of Code (GWoC)
- Finalist - Hack The Tank 3.0
- Finalist - DotSlash 9.0
Relevant Courses
Higher Secondary (12th)
Science Stream (PCM)
Mount Index International School
Completed Higher Secondary Education with focus on Physics, Chemistry, and Mathematics.
Key Highlights
- JEE Advanced : Qualified
- JEE Mains Percentile: 99.51%ile
- Boards : 90.4%
Relevant Courses
Secondary (10th)
A.S.M.P. Public School
Completed Secondary Education with good academic performance.
Key Highlights
- Boards : 95.2%
- Excellent Academic Record
- Active in Sports and Cultural Activities
Relevant Courses
Work Experience
My professional journey and the roles that shaped my skills.
Summer Intern
Motilal Nehru National Institute of Technology (MNNIT) Allahabad
Worked on an AI/ML-based Crowd Management System for Mahakumbh 2025. Built real-time detection and tracking models to enhance public safety during large-scale gatherings.
Key Responsibilities
- Developed and optimized object detection pipelines using YOLO models for real-time crowd monitoring.
- Implemented Kalman Filter-based motion tracking, improving tracking stability during occlusions.
- Designed and tested person re-identification modules using Siamese networks and face-recognition models for long-term tracking.
- Collaborated with researchers and project mentors to refine system architecture and improve detection accuracy and performance.
Featured Projects
A collection of things I've built — from side projects to production apps.
Eco-Route AI
A multi-stakeholder environmental intelligence platform. It empowers governments to plan sustainably, forest officers to protect wildlife efficiently, and citizens to travel safely and responsibly.
The Eco-Friendly Road Planning and Forest Fire Management Platform is a comprehensive web application designed to provide a sustainable and environmentally conscious approach to road planning and forest fire management. The platform utilizes cutting-edge technologies, including Next.js, React, and Firebase, to deliver a seamless and intuitive user experience.
Key Features
- Eco-friendly road planning: Plans routes using environmental data (forests, wildlife habitats) to reduce construction impact.
- Forest fire management: Uses satellite imagery and ML to detect and predict fires for faster response.
- Data analytics: Dashboard provides insights on road usage, forest health, and environmental impact for informed decisions.
- Image compression: Reduces image size for easier uploading and sharing.
- Authentication: Secure system ensures safe user access and protects sensitive data.
TransitFlow
Revolutionizing Urban Mobility - Seamless transportation solutions for campuses and cities
A comprehensive urban mobility platform for students and city commuters integrating smart parking, carpooling, route optimization, bus info, and future transport technologies.
Key Features
- Real-time smart parking availability with booking system
- Carpooling platform with ride management, messaging, and ratings
- Interactive route calculator & fare optimizer with multi-modal comparison
- Bus schedules and stop details with search functionality
- Future transport hub featuring Hyperloop, RoboTaxi, AeroTaxi, and Personal Drone
Enfinity Vision
State-of-the-art Off-road Semantic Segmentation using DINOv2 + DPT + LoRA
A deep learning-based semantic segmentation system designed for unstructured off-road environments. The model combines DINOv2 ViT-L backbone, DPT decoder, and LoRA fine-tuning to achieve high accuracy in terrain understanding across 10 classes. It enables robust scene parsing for applications like autonomous navigation and environmental perception.
Key Features
- Self-supervised DINOv2 ViT-L backbone for rich feature extraction
- DPT decoder for dense multi-scale segmentation
- LoRA-based parameter-efficient fine-tuning (~300K params)
- Hybrid architecture with CNN branch for fine spatial details
- Achieves ~66.7% mIoU and ~87% pixel accuracy on validation
- Handles 10 terrain classes including sky, grass, rocks, and trees
- Test-time augmentation (TTA) for improved inference accuracy
- Optimized training with two-phase strategy (freeze + LoRA tuning)
Chhatra Sansad
Official website for Chhatra Sansad 2026, a youth parliamentary simulation event organized by Think India, SVNIT Surat.
Developed a full-stack web platform for Chhatra Sansad to manage delegate participation and event operations. The platform includes delegate authentication, real-time announcements, admin controls, and interactive features to streamline the overall event experience.
Key Features
- Delegate login and authentication system
- Admin panel for managing delegates and content
- Real-time announcements system
- Digital ID card generation
- Event resources and document access
- Responsive UI for all devices
Genesis - AI Generated Image Detector
Detecting AI-Generated Images with Explainable Deep Learning
A full-stack AI-powered platform that accurately identifies whether an image is AI-generated or real using a ConvNeXt-based deep learning model, complete with explainable Grad-CAM visualizations, authentication, and analysis history tracking.
Key Features
- High-accuracy AI image detection using ConvNeXt architecture
- Grad-CAM heatmaps for explainable AI predictions
- Secure user authentication and profile-based history tracking
- Image upload, analysis dashboard, and result visualization
- Cloud-based image storage and scalable inference pipeline
PropValue AI
AI-powered real estate valuation — predicting accurate property prices in Surat
A machine learning–based web application that predicts house prices in Surat using historical property data, advanced data cleaning, feature engineering, and regression models. The platform provides instant price estimates through an intuitive frontend backed by a robust ML-driven API.
Key Features
- Accurate house price prediction using trained ML regression models
- Extensive data cleaning and feature engineering on real-world real estate data
- Support for key property parameters like location, BHK, area, furnishing, and transaction type
- FastAPI-based backend for real-time predictions
- Interactive React frontend for seamless user experience
JoyJuncture
Full-stack gamified platform for games, events & community experiences
A comprehensive full-stack platform for a board & card game company combining e-commerce, event management, online mini-games, blogging, and a reward system. Features rich animations, real-time integrations, and a seamless user experience.
Key Features
- Full e-commerce system with cart, filters & Razorpay payments
- Event management with booking, calendar view & Google Calendar sync
- Online mini-games (2048, Sudoku, riddles, movie quiz, spin wheel) with rewards
- Coin-based wallet system with real discount redemption
- Community blog with voting, moderation & rich content
- Experience booking system with multi-step forms
- Secure authentication with OTP, Google Sign-In & session control
- Admin dashboard with full CRUD, analytics & moderation tools
- Smooth UI with GSAP, Framer Motion, parallax & animations
Portfolio Website
Personal portfolio website showcasing skills, projects, and achievements
A modern, responsive portfolio website built with Next.js and Framer Motion animations.
Key Features
- Responsive design for all devices
- Smooth animations and transitions
- Dark theme
- Interactive project showcase
Plant Disease Detection System
Deep learning-based agricultural platform with disease prediction & marketplace
A full-stack application that uses deep learning (ResNet18) to detect plant diseases from leaf images. It identifies plant type, predicts diseases with confidence scores, and provides treatment recommendations. The platform also includes an e-commerce marketplace, community blog, and user authentication.
Key Features
- Plant disease detection using fine-tuned ResNet18 models (14 plant types)
- Image upload with prediction, confidence score & treatment suggestions
- Transfer learning using PyTorch on Kaggle dataset
- Separate models per plant with optimized loading & caching
- Integrated e-commerce marketplace for agricultural products
- Community blog platform for farmers
- Secure authentication system with hashed passwords
- REST API with FastAPI and full-stack integration
Flight Booking System
Full-stack flight booking platform with authentication & admin control
A full-stack flight booking system that enables users to search flights, manage bookings, and authenticate securely. Includes an admin panel for managing flights and monitoring bookings, ensuring a smooth and efficient booking experience.
Key Features
- JWT-based user authentication with secure login & registration
- Flight search by origin, destination, and date
- Booking management for users (view & manage bookings)
- Admin dashboard for flight CRUD operations and booking oversight
- Real-time data validation using Pydantic
- Secure password hashing with passlib
- CORS-enabled backend for secure frontend communication
ExamMaster
JEE Advanced mock test & study resource platform
A responsive web platform designed to organize and streamline exam preparation. It allows users to browse subject-wise study materials like books and notes, offering a clean and efficient way to access learning resources.
Key Features
- Subject-wise categorization of books and notes
- Responsive and modern UI for smooth user experience
- Dynamic content generation using JavaScript
- Simple and intuitive navigation for quick access
- Backend routing and request handling with Express.js
- Database integration using MongoDB with Mongoose
Hackathons & Competitions
Building under pressure, competing with the best.
PowerMind Hackathon
Adani Group
Built an AI-powered energy analytics platform that predicts electricity consumption using machine learning and provides smart insights through an interactive dashboard and Gemini AI chatbot.
Enfinity
CEV SVNIT
Built a high-performance off-road semantic segmentation system for autonomous navigation using deep learning. Designed a hybrid architecture with DINOv2 backbone and DPT decoder, achieving strong terrain understanding across diverse environments. Optimized model using LoRA fine-tuning and multi-scale feature fusion for improved accuracy and generalization.
DotSlash 9.0
ACM / ASHINE SVNIT
Built EcoRoute AI, a multi-stakeholder environmental intelligence platform for sustainable route planning and forest monitoring. Implemented eco-friendly route optimization using OSRM API and PyTorch models to minimize wildlife risk and ecological disruption. Integrated satellite-based forest fire detection and interactive dashboards for real-time environmental insights.
Hack The Tank 3.0
GDGC SVNIT
Built an AI-powered returns automation SaaS solution for Openleaf, designed to streamline product return and exchange workflows for D2C brands. Integrated Gemini AI to analyze return requests and assist in decision-making, reducing manual review effort. Developed a scalable full-stack system using Next.js, Firebase, and a custom NPM SDK for seamless integration into existing platforms.
Google Winter of Code
GDGC SVNIT
Built JoyJuncture, a full-stack digital ecosystem for a board & card game company, integrating e-commerce, event management, online games, and a gamified rewards system. Developed 60+ API endpoints and integrated multiple services including Razorpay, Firebase, Google Maps, and Calendar APIs. Implemented real-time features, secure authentication, and an admin dashboard for complete platform management.
Web Wonders
Nexus SVNIT
Built TransitFlow, a one-stop transportation platform addressing urban mobility challenges. Designed interactive modules to visualize future transport systems and developed a responsive, modular frontend using React. Focused on usability, scalability, and seamless integration of multiple transport features into a unified platform.
Technical Skills
Programming Languages
C/C++
JavaScript
Python
Web Development
Firebase
Node.js
React.js
ExpressJS
NextJS
Data Structures & Algorithms
Trees & Graphs
Arrays & Strings
Backtracking
Dynamic Programming
Linked Lists, Stacks & Queues
Tools & Technologies
Postman
Linux/Unix
Git & GitHub
Core Subjects
Object Oriented Programming
Computer Organisation
Database Management
Artificial Intelligence
Computer Networks
ML Frameworks & Libraries
Scikit-Learn
NumPy
Matplotlib & Seaborn
PyTorch
Pandas
OpenCV
...
Total Visitors