building apps cuz it's fun

Paarangat Rai Sharma  • Peerlist

RailKavach is an AI-driven railway safety system designed to prevent train collisions with obstacles, stray animals, and humans. It uses real-time object detection, multi-modal sensor fusion (CCTV, LiDAR, Infrared), and automated braking to reduce railway accidents. How It Works: AI-Powered Detection: YOLOv5 + VGG16 models detect obstacles in real-time. Multi-Sensor Fusion: LiDAR, CCTV, and infrared cameras ensure high-accuracy detection, even in fog, rain, or low-light conditions. Automated Braking: If the system detects an obstruction and no manual response occurs, RailKavach automatically stops the train. Key Features: Scalable – Works with existing railway infrastructure. Edge Computing – Low-latency AI processing. Cost-Effective – Uses affordable sensor tech. Challenges & Future Plans: Integration: While the frontend and AI model were built, seamless real-time integration remains a priority. Data Limitations: We relied on an image dataset due to the lack of Indian Railways live footage. Access to real-time data would significantly enhance detection accuracy. 🏆 Award: 1st Place – Open Innovation Track, Hackurate, Zenevia 2024. 👨💻 Team Hawks: Anusha Pundir, Sumukh Chhabra, Yash Padam, Yash Nautiyal, Paarangat Rai Sharma.

RailKavach | AI Powered Train Collision Preve

ResQ is a real time disaster response platform designed to help citizens report emergencies instantly while giving authorities an intelligent system to verify, classify, and respond using multimodal AI. Our team built it to solve a major issue during disasters: false reports, slow verification, and fragmented data that delay critical action. ResQ provides structured incident reporting, automated validation, and an authority dashboard for rapid decision making. As part of the core engineering team, I led the mobile development, AI pipeline integration, and the underlying backend flows. I built 7 full production level screens across two user profiles: • Citizen App: Onboarding, OTP email login, incident reporting, location tagging, severity insights, nearby feed (for alerts by users in nearby locations within 5km radius) • Authority App: Dashboard, incident review panel with AI scores, map overlays, automatic resource allocation I designed these screens end to end using React Native, focusing on clean UX, fast interactions, and smooth transitions between citizen submissions and authority decisions. On the AI side, my teammates trained models like EfficientNet, MobileNet, and YOLOv5, while I integrated the entire AI pipeline into the app and built the Gemini verification layer. For the backend, I implemented Supabase based OTP authentication, role separation for citizen and authority users, incident storage with geolocation, and real time syncing between public reports and authority decisions. 🏆 ResQ earned 2nd place out of 147 teams in our university's Project Showcase Skills demonstrated: React Native, Supabase (Auth, DB, Storage), AI model integration, YOLOv5, EfficientNet, Gemini API, REST APIs, mobile UX, real time architecture, teamwork, and end to end product development.

ResQ • AI Powered Disaster Response Mobile Ap

Talx is an alternative to Wispr Flow. It's an open-source voice-to-text dictation tool for macOS. Speak anywhere on your system and get polished, context-aware text instantly pasted into any app. Built with a Rust (Tauri v2) backend for native performance and a React/TypeScript frontend as a minimal floating widget. Uses Groq Whisper for speech recognition and LLM-powered transcript polishing to clean up filler words, fix grammar, and format output naturally. A bring-your-own-API-key alternative to subscription dictation apps like Wispr Flow, designed for full control over the experience and cost.

Talx • AI-powered voice dictation for macOS

TrajectoryAI + ProfessorX is an AI-powered career mapping tool designed to help students and professionals navigate their career choices through interactive AI mentorship. Our platform solves the problem of career confusion by providing personalized roadmaps, real-time job market insights, and an AI chatbot mentor capable of voice-based consultations. Key Features: AI-Powered Career Pathway Mapping Speech-to-Text & Voice AI Mentor (Powered by Gemini’s API) Real-Time Job Market Insights & Skill Recommendations Gamified User Experience with interactive roadmap visualization Scalable & Accessible, designed for students and professionals across industries My Role: I spearheaded the front-end development for ProfessorX, an AI chatbot that allows users to ask career-related questions via voice or text, and receive personalized guidance through a visually engaging roadmap. The interface was designed to represent a decision tree, dynamically branching into tailored career paths based on user input. On answering a question, based on the answer, the node would move forward and present you a personalized roadmap made just for you. Hackathon Achievements: Built in 26 hours at IIT Delhi’s TRYST Tech Fest, 2025. Recognized for innovation & real-world impact. Overcame last-minute team reshuffle to deliver a winning solution. 🏆 1st Runner-Up, IIT Delhi Cube Innovators Hackathon

TrajectoryAI + ProfessorX | AI-powered Career

AI-powered vision board platform that generates photorealistic images of your goals and aspirations using your own face and detailed scene descriptions. Building a full-stack SaaS solo - database schema with RLS, API layer, AI integration, payment processing, and responsive frontend Designed an 8-step Vision Studio wizard that walks users through category selection, style customization, face upload, background composition, and AI image generation with smooth step transitions Engineered a structured prompt system that translates user inputs into multi-block prompts with category-aware defaults, face identity preservation, camera parameter mapping, and relationship-specific framing across 10 focus areas Integrated Google Gemini 2.5 Flash Image for image generation with input sanitization against prompt injection, magic byte file validation for uploads, and face reference handling via inline base64 Built a credit-based billing system with Dodo Payments - webhook-driven subscription lifecycle with idempotency, atomic credit consumption via database functions, and grace periods for canceled plans Implemented Cloudflare R2 storage with signed URLs, retry-with-backoff uploads, and a client-side Zustand cache with TTL to minimize redundant URL signing Secured the platform with Row Level Security on all tables, an API handler factory with automatic auth and Zod validation, sliding window rate limiting, and security headers Tech: Next.js 16 · React 19 · TypeScript · Tailwind · Supabase · Google Gemini API · Cloudflare R2 · Dodo Payments · Zustand · Framer Motion · Zod