O

open-source

An open-source, modular public transport routing engine built with Python 3.13+. It abstracts external transit APIs behind a resilient service layer to compute efficient multi-modal routes with integrated caching and strict compliance.

Offline-first, CPU-first AI reader companion that narrates and analyzes stories locally using Ollama and structured feedback JSON.

Offline-first lecture transcription and structured study data extraction CLI/PWA.

Uncover Truth is a full-stack, AGPLv3-licensed logic puzzle web application built using a modern JavaScript monorepo architecture. The project combines a React-based frontend, a Node.js/Express backend, and shared game logic to deliver an interactive puzzle-solving experience focused on reasoning and deduction.

The system is designed with production-grade engineering practices, including automated testing, code quality enforcement, security scanning, and CI/CD compliance pipelines using GitLab.

⚙️ Key Features 🧩 Interactive logic puzzle gameplay with shared puzzle engine ⚛️ Modern frontend built with React and Vite 🖥️ Lightweight Node.js + Express backend 📦 Monorepo structure with shared utilities between client and server 🧪 Automated testing using Vitest with coverage reporting 🔍 Code quality enforcement via ESLint, Prettier, and Knip 🔐 Security scanning using Gitleaks and npm audit 📊 CI/CD pipeline with compliance checks (GitLab-ready) 📄 Open-source governance: LICENSE, CONTRIBUTING, SECURITY, CODE_OF_CONDUCT 🏗️ Architecture client/ → React frontend (Vite) server/ → Express backend API shared/ → Shared puzzle logic used by both client and server test/ → Unit and integration tests CI/CD → GitLab pipeline with lint, security, test, and compliance stages 🚀 Purpose

This project is designed to demonstrate real-world full-stack engineering practices, focusing on:

Clean architecture in monorepos Secure and compliant development workflows Automated testing and coverage enforcement Production-ready CI/CD pipelines Open-source governance standards

CivicPlus is a citizen grievance management platform built with React, TypeScript and Node.js.

LocalSlate is an offline-first, CPU-optimized AI incident processing system that converts unstructured text, audio, and voice reports into structured actionable data using a FastAPI backend, local AI workflow, and MySQL persistence.

Memento AI – A CPU-first, offline AI assistant that transforms documents, images, audio, video, and text into structured knowledge using local LLMs, SQLite, and Retrieval-Augmented Generation (RAG), with

Open-source crowdsourced wildfire and smoke tracking dashboard built for communities.

Offline Investigation Intelligence Platform

StoryMirror CLI is a CPU-first command-line application that transforms unstructured stories into meaningful insights through local AI processing. It provides story analysis, multilingual text-to-speech (TTS), runtime performance metrics, and customizable terminal themes, all while running on a local FastAPI backend.

Designed for offline and resource-efficient environments, StoryMirror performs inference on the CPU without requiring a dedicated GPU, making it suitable for developers, students, and writers who need an accessible AI-powered storytelling tool.

Key Features 📖 Analyze stories from text or files 🎙️ Generate multilingual text-to-speech narration 🎨 Customize the CLI with multiple themes 📊 Monitor CPU usage, memory, and inference performance ⚡ Run entirely on a local FastAPI backend 💻 CPU-first design with no GPU dependency 🌐 Supports offline or local deployment 🚀 Simple command-line interface for quick workflows

AgriDoc is a CPU-first, offline-first web app that helps India's farmers digitize and manage their critical agricultural documents — Pattadar Passbooks, Adangal records, PM-KISAN certificates, Rythu Bandhu slips, and more.

EcoSort AI is an offline-first, CPU-powered web application that transforms unstructured garbage images into structured waste intelligence. Using local AI inference, it identifies waste categories, generates structured JSON reports, and stores results in SQLite for analysis and retrieval. The system works without an internet connection, making it suitable for smart cities, municipalities, campuses, and remote environment

SafeRoute AI is a machine-learning-powered road safety platform that predicts accident blackspots, analyzes route risk, and recommends safer travel routes using FastAPI, Next.js, SQLite, and XGBoost.

A full-stack, AI-powered health twin simulator with a built-in interactive knowledge base and brain games module.

AI-powered decision simulation platform — multi-expert adversarial debate with real-time streaming, structured scoring, and AI-synthesised verdicts. Built for the Swecha Internship Hackathon.

Khetika — A multilingual AI-powered farming assistant that helps Indian farmers diagnose crop diseases, plan weekly tasks, and get instant advice in their native language.

This is the Bilingual AI powered chatbot that I developed as a part of the Summer of AI internship task.

AI-powered chatbot application with React frontend and Node.js backend for interactive conversations

Ruchulu Lens AI

Ruchulu Lens AI is a cross-platform AI-powered application that celebrates India's regional food diversity. The app offers:

Regional Food Explorer – Shows local dishes based on user location, revealing region-specific ingredients, cooking styles, and preparation. Lens Feature (Food Recognition) – Users can upload or capture an image of a dish, and the AI will it and provide ingredients, recipe steps, and cultural context. AI Culinary Assistant – Interactive assistant to answer cooking-related questions, suggest dietary plans, offer substitutions, and guide users through cooking.