Developed a Travel Companion Application utilizing Google Maps API, Geolocation, and advanced React/Next.js and TypeScript practices, leading to seamless user experience and efficient data handling. Gained proficiency in using Google Cloud Platform (GCP) for Google Maps API setup, and account/billing management.
Used Material-UI for creating the interface and leveraged Travel Advisor API to fetch location-specific data on restaurants and hotels for a highly personalized user experience, including filtering of locations based on ratings.
Ensured app security and maintainability with environment variables, hooks, refs, and a well-organized folder structure, resulting in a highly scalable and deployable project on Vercel.
Developed and deployed a YouTube clone application on Vercel in React/Next.js and TypeScript, leveraging Material UI for a visually appealing and responsive design, and using RapidAPI to access YouTube v3 API endpoints for fetching video and channel data.
Acquired expertise in React functional components and their reusability, efficient file and folder structure, and mastery of Material UI, enabling the ability to build any API-driven web application.
Ensured optimal responsiveness across devices using well-crafted media queries, and utilized RapidAPI Vscode extension for API testing, demonstrating proficiency in creating a seamless, fully-functional video browsing experience.
Developed and deployed a full-stack chat application with authentication, Twilio SMS notifications, and a wide array of features, including direct/group chats, emojis, reactions, GIF support, and message editing, using JavaScript/React/Node.js/Express, and Stream API.
Implemented advanced React best practices, custom hooks, and React Context API to create a responsive, and scalable chat interface; and acquired skills to create fully custom chat applications by doing this project.
Seamlessly integrated Twilio for real-time SMS notifications for offline users and employed Vercel to deploy the frontend application and Render to deploy the Express server, showcasing proficiency in diverse technology stacks and third-party APIs for a complete, mobile-responsive chat experience.
Developed and deployed a fully-responsive, modern full-stack e-commerce application on Vercel using React, Next.js, and TypeScript with integrated Stripe payments, advanced cart functionalities, and dynamic content management using Sanity for seamless content updates.
Implemented engaging UI features such as hover animations, dynamic product detail pages, and animated recommendations, ensuring an optimal shopping experience across all devices.
Implemented Stripe integration for secure payment processing, shipping method selection, and checkout, resulting in a seamless and efficient customer purchasing experience.
Developed a comprehensive sorting algorithm visualizer website featuring over 30 unique algorithms, built using React, Redux, and Styled Components to facilitate learning and understanding of various sorting concepts.
Integrated Material UI, Framer Motion, react-animate-height, and react-code-blocks to enhance user experience with interactive icons, sliders, responsive sidebars, modals, and collapsible list elements.
Utilized Create React App for initial project template, allowing for rapid development and customization of the educational tool to provide custom delay times and input array sizes for an optimized learning experience.
Developed a group video chat application with support for multiple participants, screen sharing, private and group messaging, admin controls, and polling, using Agora’s chat builder for scalability and rapid development.
Enhanced existing Agora codebase with additional features such as a polling system, showcasing the ability to work effectively in large codebase and integrate new functionalities.
Implemented user-friendly features such as meeting creation and sharing, admin controls for muting and managing participants, and options for private or group chats, promoting an engaging user experience.
Deployed the frontend on Vercel and the backend on Heroku.
Transformed a Figma design into a fully responsive website with modern UI/UX using React, Next.js, and TypeScript, showcasing proficiency in design-to-development implementation and seamless deployment.
Leveraged React functional components and reusable code, adhering to best practices in file and folder structure, resulting in a well-organized and maintainable web application.
Mastered essential CSS properties for flex and grid layouts, applied BEM methodology for scalable styling, and integrated soft animations and complex gradients, along with carefully crafted media queries for optimal responsiveness across various devices.
Developed on online bookstore application using Django and HTML/CSS/JavaScript, featuring secure user authentication, account activation, profile editing, and password management with automated email notifications to enhance user experience and security.
Implemented advanced search functionality, allowing users to filter books by category, author/publisher, ratings, and price, alongside a seamless ordering process, including promotion code integration, order summary, and reordering options.
Enabled admin capabilities to add and manage promotions, with automatic email notifications to opted-in users, while incorporating inventory checks for reordering and ensuring promotion validity, resulting in a comprehensive and user-friendly e-commerce platform.
Built PHP/MySQL program for Chandigarh, India police department
Assign duties at various stations without repetition between days
Jupyter Notebooks
Mish activation function is tested for transfer learning. Here mish is used only in the last fully-connected layers of a pretrained Resnet50 model. I test the activation function of CIFAR10, CIFAR100 using three different learning rate values. I found that Mish gave better results than ReLU. notebook, paper.
Multi Sample Dropout is implemented and tested on CIFAR-100 using cyclic learning. My losses converged 4x faster when using num_samples=8 than using simple dropout. notebook, paper.
Data Augmentation in Computer Vision
Notebook implementing single image data augmentation techniques using just Python notebook.
Summarizing Leslie N. Smith’s research in cyclic learning and hyper-parameter setting techniques. notebook
A disciplined approach to neural network hyper-parameters: Part 1 learning rate, batch size, momentum, and weight decay paper.
Super-Convergence: Very Fast Training of Neural Networks Using Learning Rates paper.
Exploring loss function topology with cyclical learning rates paper.
Cyclical Learning Rates for Training Neural Networks paper.
Photorealistic Style Transfer. Implementation of the High-Resolution Network for Photorealistic Style Transfer paper. notebook, paper.
Weight Standardization is implemented and tested using cyclic learning. I find that it does not work well with cyclic learning when using CIFAR-10. notebook, paper.
PyTorch computer vision tutorial. AlexNet with tips and checks on how to train CNNs. The following things are included: notebook.
Dataloader creation
Plotting dataloader results
Weight Initialization
Simple training loop
Overfitting a mini-batch
Waste Segregation using trashnetgithub. Contains the code to train models for trashnet and then export them using ONNX. It was part of a bigger project where we ran these models on Rasberry Pi, which controlled wooden planks to classify the waste into different categories (code for rasberry pi not included here).
