Perplexus AI Full-Stack Answer Engine 🧠⚡

A modern AI-powered conversational search and research engine combining LangChain agent orchestration, live Tavily web retrieval, Pinecone vector RAG, multimodal document/vision analysis, and real-time streaming over WebSockets.

---

🌟 Key Features

• Multi-Engine AI Synthesis: Queries orchestrated through LangChain leveraging Mistral AI, OpenAI, and Google Gemini for deep reasoning and synthesis.
• Hybrid Vector RAG: Semantic vector retrieval using Pinecone DB, Recursive Character Text Chunking, and high-dimensional embeddings.
• Live Internet Grounding: Real-time web retrieval using Tavily Search API for up-to-the-minute web citations and source transparency.
• Multimodal Document & Vision Parsing:
  • Image comprehension via Google Gemini (gemini-2.5-flash).
  • Document ingestion for PDF (pdf-parse) and DOCX (mammoth) files.
  • Image hosting and CDN asset management via ImageKit.
• Real-Time Token Streaming: Low-latency bidirectional socket streaming with typewriter markdown rendering, code highlighting, and auto-generated thread titling.
• Perplexus "Scholar's Parchment" UI:
  • Calibrated warm parchment aesthetic (#faf8f5 canvas, #016a71 deep teal accents, #27251e ink typography, and custom editorial serif logo mark).
  • Voice query integration via Web Speech API and dynamic model selector.

---

🛠️ Tech Stack

Frontend

• Framework: React 19, Vite 7, React Router 7
• State Management: Redux Toolkit, React-Redux
• Styling: Tailwind CSS v4 (@tailwindcss/vite), Framer Motion
• Markdown & UI: React-Markdown, Remark-GFM, Lucide React, React Hot Toast
• Networking & Real-Time: Socket.io-client, Axios

Backend

• Server Runtime: Node.js, Express v5
• AI & Agent Orchestration: LangChain, @langchain/google-genai, @langchain/mistralai, @langchain/openai, @langchain/textsplitters, @tavily/core, Zod
• Vector Database: Pinecone Database (@pinecone-database/pinecone)
• Document & Media Processors: PDF-Parse, Mammoth (DOCX), Multer, ImageKit
• Database & Auth: MongoDB, Mongoose v9, JWT, Bcryptjs, Nodemailer, Cookie-Parser

---

📁 Repository Architecture

04 - Perplexity/
├── Backend/
│   ├── src/
│   │   ├── config/          # DB & third-party service configs
│   │   ├── controllers/     # Auth & Chat controllers
│   │   ├── middlewares/     # JWT authentication & Multer uploads
│   │   ├── models/          # User, Chat, Message Mongoose models
│   │   ├── routes/          # Express API routes
│   │   ├── services/        # AI (Gemini/Mistral), RAG (Pinecone), Tavily search, Mail
│   │   ├── sockets/         # Socket.io live streaming handlers
│   │   └── validators/      # Express-validator schemas
│   ├── server.js            # Server entry point
│   └── package.json
└── Frontend/
    ├── src/
    │   ├── app/             # Redux store & router setup
    │   ├── features/
    │   │   ├── auth/        # Auth slice, login, register, protected wrapper
    │   │   └── chat/        # Chat slice, socket service, typewriter hook, feed, sidebar
    │   ├── main.jsx
    │   └── app/index.css
    └── package.json

---

🚀 Getting Started

Prerequisites

• Node.js (v18+)
• MongoDB connection URI
• Pinecone API Key & Index
• API Keys: Gemini, Mistral, Tavily, ImageKit

1. Configure & Run Backend

cd Backend
npm install

Create .env in Backend/:

PORT=3000
MONGODB_URI=your_mongodb_connection_string
JWT_SECRET=your_jwt_secret

GEMINI_API_KEY=your_gemini_api_key
MISTRAL_API_KEY=your_mistral_api_key
TAVILY_API_KEY=your_tavily_api_key

PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX=your_pinecone_index

IMAGEKIT_PUBLIC_KEY=your_imagekit_public_key
IMAGEKIT_PRIVATE_KEY=your_imagekit_private_key
IMAGEKIT_URL_ENDPOINT=your_imagekit_url_endpoint

MAIL_USER=your_email@example.com
MAIL_PASS=your_email_password

Start backend:

npm run dev

2. Configure & Run Frontend

cd ../Frontend
npm install
npm run dev

---

📡 API & Socket Overview

Channel / Route	Type	Description
/api/auth/login	HTTP POST	Authenticate user session
/api/chats	HTTP GET/POST	Fetch user conversations / Create thread
join_room	Socket Event	Join real-time streaming thread
send_message	Socket Event	Dispatch prompt for live RAG + LLM streaming
receive_chunk	Socket Event	Receive streamed tokens with markdown chunks

---

📄 License

ISC License. Built for portfolio showcase.