Skip to main content

🛠️ Projects

Fuelled by a burning curiosity about what makes software tick, and the itch to unravel the mysteries under the hood, born from years of being a software aficionado, I ventured into the world of software development. For me, crafting software isn't just a skill; it's an immersive way of learning, a journey where lines of code weave together to demystify the intricacies of technology.

 

Traffic​

A deep learning solution for real-time traffic sign classification using convolutional neural networks. The project implements a CNN-based classifier to identify 43 different categories of German traffic signs. It explores hierarchical feature extraction, comparing a basic single convolutional layer approach with an improved multi-layer architecture for significantly better accuracy.

Nim AI​

Train an AI to master the game of Nim using Q-learning reinforcement learning. The AI learns the optimal strategy by playing thousands of games against itself, updating Q-values using the standard Q-learning update rule. After sufficient training, it exploits mathematical properties of Nim to make moves that lead to victories.

Shopping​

A K-Nearest Neighbor machine learning tool designed to predict whether an online shopping session will result in a purchase. By analyzing various user behavior metrics—such as page visits, duration, bounce rates, and exit rates—the system identifies patterns that correlate with revenue generation.

Music Teacher​

A multi-agent AI music tutor that analyzes real-time MIDI input to teach music theory. Orchestrated with LangGraph and powered by Google Gemini, it provides instant feedback on chords and scales, bridging the gap between physical instrument practice and theoretical understanding.

RAG Microservice​

A Retrieval-Augmented Generation (RAG) microservice built with FastAPI and LangChain. It combines vector search (FAISS) with open-source LLMs to answer queries based on custom documents, reinforced by a security layer that detects and blocks prompt injections and malicious inputs.

Orderbook​

A high-performance multi-symbol orderbook service engineered with Node.js, TypeScript, and PostgreSQL. It supports both limit and market orders, featuring an automatic trade matching engine that efficiently processes and settles transactions in real-time.

API Schema Manager​

A robust schema management service designed for JSON and YAML OpenAPI specifications. It features comprehensive version control, supports hierarchical application/service structures, and ensures reliable file persistence, streamlining the lifecycle of API definitions.

PageRank​

A Python implementation of the PageRank algorithm, the core technology behind Google Search. It models the web as a graph of nodes and edges, using Markov Chains and iterative convergence (via both sampling and algebraic iteration) to calculate the probability-based importance of each page.

Minesweeper AI​

An intelligent Minesweeper agent that combines a pygame graphical interface with a propositional logic engine. The AI uses logical inference and constraint satisfaction to strictly deduce safe moves and mine locations, making probabilistic guesses only when absolutely necessary.

Knights and Knaves​

An AI-powered model-checking solver for the Knights and Knaves logic puzzle. The algorithm applies propositional logic to determine the truthfulness of characters, navigating through complex logical paradoxes to find the unique solution.

Tic-Tac-Toe​

A graphical Tic-Tac-Toe game implementing the Minimax adversarial search algorithm with Alpha-Beta pruning for optimal gameplay. The project demonstrates the application of game theory concepts in a classic zero-sum game setting.

Yoga Pose Recommender: AI-Powered Vector Search for Yoga Poses​

An AI-powered semantic search system that helps users explore yoga poses. By leveraging Gemini (LLM) to generate rich, context-aware descriptions and create high-dimensional vector embeddings, the project transforms an existing Hugging Face dataset into a deeply intelligent and intuitive search experience. With LangChain JS orchestrating data processing and Firestore serving as a high-performance vector database, users can retrieve the most relevant yoga poses through natural language queries. Additionally, Google Cloud’s Text-to-Speech technology enhances accessibility by providing immersive audio descriptions. Built with Node.js, this system seamlessly integrates AI-driven search, vector embeddings, and real-time query processing to redefine yoga exploration.

Command Line Generative AI​

A true computer geek never leaves the terminal and tries to achieve everything via the command line. Built by the geek, for the geeks, a command line generative AI bot that can be used directly from the terminal without the need for going to a browser. Utilizes Google's Large Language Model(LLM), Gemini Pro.

Source Crowd​

Designed and built a web application using Django, where people post links to the webpages that they find helpful, and others can vote based on the quality. They can also add a title and description for that link, based on which others can search for that webpage. It's a crossover of Google Search and Reddit.

Emoji Memory Game​

An iOS game application, built using Swift and SwiftUI, where users have to find the matching emoji, hidden behind the cards, and appears on tap. Features include play again, choice of level, and choice of emoji.

Network​

A full-stack social networking web app like Twitter(now X), for making, deleting, and editing posts. Using Fetch API, implemented the following features: like and unlike posts, and follow other users.

Mail​

Front-end for an email client that makes API calls to send and receive emails, reply and archive emails, and view different mailboxes.

Commerce​

An eBay-like e-commerce auction web app that allowed users to post auction listings, place bids on listings, comment on those listings, and add listings to a “watchlist.”

Wiki​

Wikipedia-like online encyclopedia, where anybody can create a new page and also edit it. Functionalities also include searching for pages based on their title.

Front-end for Google Search, Google Image Search, and Google Advanced Search, which will redirect users to the actual Google search results page.