Full Stack & AI Engineer building real-time AI apps with Next.js, FastAPI, NestJS, and Computer Vision.

leetcode

youtube

github

ABHISHEK ABHISHEK  • Peerlist

<h3>AI FAQ Assistant</h3><p>An AI-powered FAQ chatbot built with <strong>Next.js 15</strong>, <strong>NestJS</strong>, <strong>MongoDB</strong>, and <strong>Google Gemini AI</strong>.</p><h3>Architecture</h3><p>The project is structured into two main parts:</p><ul><li><strong>Frontend</strong>: Developed using <strong>Next.js 15</strong> (with React 19 and TypeScript), featuring an App Router for pages and layouts, reusable components like ChatMessage, ChatInput, Sidebar, and Header, and an API client in lib/api.ts.</li><li><strong>Backend</strong>: Built with <strong>NestJS</strong> (TypeScript), containing modules for chat functionality (controller, service, schema, DTOs), Gemini AI integration, and core application setup.</li></ul><p>A docker-compose.yml file is provided for containerization.</p><h3>Prerequisites</h3><ul><li>Node.js 20+</li><li>npm or yarn</li><li>MongoDB Atlas account (or local MongoDB)</li><li>Google Gemini API key</li></ul><h3>Setup Instructions</h3><p><strong>Option A: Run Without Docker (Manual)</strong></p><ul><li><strong>Backend</strong>: Navigate to the backend directory, run npm install, and then npm run start:dev. The backend will run on http://localhost:3001.</li><li><strong>Frontend</strong>: Navigate to the frontend directory, run npm install --legacy-peer-deps, and then npm run dev. The frontend will run on http://localhost:3000.</li></ul><p><strong>Option B: Run With Docker</strong></p><ul><li>From the root directory, run docker-compose up --build.</li></ul><p>The application will be accessible at http://localhost:3000 (Frontend) and http://localhost:3001 (Backend).</p><h3>API Endpoints</h3>POST/api/chat/askAsk a question (standard)POST/api/chat/ask/streamAsk with streaming (SSE)GET/api/chat/historyGet paginated historyGET/api/chat/search?q=querySearch conversationsGET/api/chat/statsGet total/today countsDELETE/api/chat/:idDelete a conversation<h3>Features</h3><h4>Core</h4><ul><li>✅ Chat interface with user/AI message bubbles</li><li>✅ AI-powered answers via Google Gemini 1.5 Flash</li><li>✅ Conversations stored in MongoDB Atlas</li><li>✅ Conversation history API with pagination</li></ul><h4>Bonus</h4><ul><li>✅ <strong>Streaming responses</strong> — real-time token-by-token display via SSE</li><li>✅ <strong>Dark mode</strong> — toggle with persistence in localStorage</li><li>✅ <strong>Conversation search</strong> — full-text search with regex fallback</li><li>✅ <strong>Docker setup</strong> — docker-compose up --build</li></ul><h3>Technical Decisions</h3><h4>Backend</h4><ul><li><strong>NestJS</strong> for structured, scalable TypeScript APIs with dependency injection</li><li><strong>Mongoose</strong> schemas with text indexes for efficient search</li><li><strong>Timestamp index</strong> for fast history queries sorted by recency</li><li><strong>Global validation pipe</strong> with class-validator for input sanitization</li><li><strong>CORS</strong> configured to allow frontend origins</li></ul><h4>Frontend</h4><ul><li><strong>Next.js 15 App Router</strong> with React 19 for the latest patterns</li><li><strong>Streaming via SSE</strong> — fetch ReadableStream for real-time display</li><li><strong>react-markdown + remark-gfm</strong> to render AI responses as rich markdown</li><li><strong>No external state library</strong> — React useState handles all local state</li><li><strong>CSS variables</strong> for instant dark/light theme switching</li></ul><h4>MongoDB Schema Design</h4><p>{<br>&nbsp;&nbsp;question: String, // Indexed (text)<br>&nbsp;&nbsp;answer: String, // Indexed (text)<br>&nbsp;&nbsp;timestamp: Date, // Indexed (desc) for sorted history<br>}<br>// Compound text index on question + answer for full-text search</p>

Ai-chatbot

<h3>About NeuroScroll</h3><p>NeuroScroll, also known as JustBlink, is an innovative AI-powered system designed for real-time human-computer interaction. It allows users to control documents and navigate interfaces using eye gestures, offering a hands-free and intelligent solution.</p><h3>Key Features:</h3><ul><li><strong>AI-Powered Eye Gesture Control:</strong> Utilizes AI and computer vision to detect blinks and other eye movements for document control.</li><li><strong>Real-time Processing:</strong> Processes live video streams to provide immediate feedback and control.</li><li><strong>Eye Aspect Ratio (EAR) Calculation:</strong> Accurately measures eye openness to differentiate between blinks and normal eye states.</li><li><strong>Blink Detection Logic:</strong> Employs a Finite State Machine for robust detection of single blinks, long blinks, and double blinks.</li><li><strong>Gesture Recognition:</strong> Translates detected eye gestures into specific actions like scrolling down, scrolling up, and clicking.</li><li><strong>Full-Stack Architecture:</strong> Integrates frontend (Next.js, React) and backend (FastAPI, WebSockets) for a seamless experience.</li><li><strong>Visual System Interface:</strong> Provides a live camera feed with facial mesh overlay and real-time AI simulation for user feedback.</li></ul><h3>Technology Stack:</h3><ul><li><strong>Frontend:</strong> Next.js (App Router), React, MediaPipe FaceMesh, WebSocket API, Tailwind CSS.</li><li><strong>Backend:</strong> FastAPI, Uvicorn, NumPy, MongoDB, Python-dotenv.</li><li><strong>AI &amp; Vision:</strong> MediaPipe FaceMesh, Eye Aspect Ratio (EAR), Blink State Machine, Gesture Classification Engine.</li></ul><p>This project is ideal for research and educational purposes, demonstrating advanced concepts in computer vision, human-computer interaction, and full-stack AI integration.</p>