🔬 ResearchAI — Multi-Agent Research Workspace

An AI-powered multi-agent research platform that transforms complex research questions into structured, evidence-backed insights using autonomous AI agents, real-time web search, RAG, vector databases, and knowledge graphs.

ResearchAI is a full-stack MERN-based AI research workspace that automates the complete research lifecycle — from research planning and source discovery to analysis, fact-checking, synthesis, knowledge visualization, comparison, deep-dive research, and PDF report generation.

---

🚀 Features

• 🤖 Multi-Agent Research — Planner, Researcher, Analyst, Fact Checker and Synthesizer agents.
• 🌐 Real-Time Web Search — Tavily-powered search with source URLs and citations.
• 📚 Document RAG — Upload PDF/TXT documents and perform contextual research using Pinecone.
• 🧠 Knowledge Graph — Neo4j-powered entity and relationship visualization.
• 💬 Research Chat — Ask questions about completed research and documents.
• 🎯 Deep Dive Research – Perform targeted follow-up research on individual findings.
• 🔎 Trace Evidence – Map claims directly to supporting and contradicting web sources and RAG documents.
• ⚔️ Debate Findings – Trigger Pro vs Counter adversarial AI agents to challenge bias and hallucination.
• ⚠️Contradiction Detection – Automatically cross-reference findings to highlight conflicting claims.
• 🔄 What Changed? – Run evolutionary analysis between past and present research to track new and retracted claims.
• 📱 Premium Responsive UI – Flawlessly optimized across mobile, tablet, and desktop viewports with discovery-first UX.
• ⚖️ Research Comparison – Compare two completed research workspaces.
• ⏳ Research Timeline — Visualize important research events chronologically.
• 🔎 Global Search — Search across workspaces and research content.
• 📄 PDF Export — Export research results into structured PDF reports.
• 🎯 Demo Workspaces — Explore pre-built research without consuming API credits.
• 🔐 JWT Authentication — Secure registration, login and protected APIs.
• 🌓 Dark/Light Mode — Responsive premium interface.

---

🏗️ System Architecture

 ┌────────────────────────┐
│ React + Vite │
│ Frontend │
└────────────┬───────────┘
│
REST API
│
┌────────────▼───────────┐
│ Node.js + Express │
│ Backend │
└────────────┬───────────┘
│
┌───────────────────────┼────────────────────────┐
│ │ │
▼ ▼ ▼
┌────────────┐ ┌────────────┐ ┌────────────┐
│ MongoDB │ │ Neo4j │ │ Pinecone │
│ Application│ │ Knowledge │ │ Vector DB │
│ Data │ │ Graph │ │ RAG │
└────────────┘ └────────────┘ └────────────┘
│
▼
┌────────────────────┐
│ Gemini AI │
│ Multi-Agent Core │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ Tavily Search │
│ Real Web Search │
└────────────────────┘

---

🔄 Multi-Agent Research Workflow

 Research Question
│
▼
┌─────────┐
│ Planner │
└────┬────┘
│
▼
┌────────────┐
│ Researcher │
└─────┬──────┘
│
▼
┌─────────┐
│ Analyst │
└────┬────┘
│
▼
┌─────────────┐
│ Fact Checker│
└──────┬──────┘
│
▼
┌─────────────┐
│ Synthesizer │
└──────┬──────┘
│
▼
Final Research Report
│
┌─────────────┼─────────────┐
▼ ▼ ▼
Web Sources Knowledge Graph Timeline
│
├──────────────► Research Chat
│
├──────────────► Deep Dive
│
└──────────────► PDF Export

---

📚 RAG Pipeline

 PDF / TXT Upload
│
▼
Text Extraction
│
▼
Chunking
│
▼
Embeddings
│
▼
Pinecone
│
▼
Semantic Retrieval
│
▼
Gemini Context
│
▼
Contextual Answer

---

🧠 Knowledge Graph Pipeline

Research Findings
│
▼
Gemini Entity Extraction
│
▼
Entities + Relationships
│
▼
Neo4j Graph Database
│
▼
Interactive Knowledge Graph

---

⚖️ Research Comparison

Users can select two completed research workspaces and compare:

• Similarities
• Differences
• Conflicting findings
• Supporting evidence
• Sources
• Overall synthesis

Workspace A ──────┐
├──► Comparison Engine
Workspace B ──────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
Similarities Differences Conflicts
│
▼
Final Synthesis

---

🔬 Deep Dive Research

Users can select any research finding and perform targeted follow-up research.

Selected Finding
│
▼
Deep Dive
│
├── Supporting Evidence
├── Latest Sources
├── Opposing Views
└── Additional Context

---

⏳ Research Timeline

ResearchAI extracts relevant dates from verified sources and organizes research events chronologically.

2022 ───── 2023 ───── 2024 ───── 2025 ───── 2026
│ │ │ │ │
Study Paper Trend Report Latest

---

🛠️ Tech Stack

Frontend

• React
• Vite
• Tailwind CSS
• Zustand
• React Router
• Lucide React
• react-force-graph-2d

Backend

• Node.js
• Express.js
• MongoDB
• Mongoose
• JWT
• bcryptjs

AI & Research

• Google Gemini
• Tavily Search API
• Multi-Agent Architecture
• Retrieval-Augmented Generation (RAG)
• Prompt Engineering

Databases

• MongoDB — Application Database
• Pinecone — Vector Database
• Neo4j — Knowledge Graph

Deployment

• Vercel

---

📁 Project Structure

ResearchAI-MultiAgent/
│
├── client/
│ ├── public/
│ ├── src/
│ │ ├── components/
│ │ │ ├── GraphPanel/
│ │ │ ├── ResearchChat/
│ │ │ ├── DeepDive/
│ │ │ └── ...
│ │ ├── pages/
│ │ │ ├── Dashboard/
│ │ │ ├── Workspace/
│ │ │ ├── CompareWorkspaces/
│ │ │ └── ...
│ │ ├── services/
│ │ ├── store/
│ │ ├── App.jsx
│ │ └── main.jsx
│ └── package.json
│
├── server/
│ ├── config/
│ ├── controllers/
│ ├── middleware/
│ ├── models/
│ ├── routes/
│ ├── services/
│ │ ├── agents/
│ │ │ ├── planner/
│ │ │ ├── researcher/
│ │ │ ├── analyst/
│ │ │ ├── factChecker/
│ │ │ └── synthesizer/
│ │ ├── ai/
│ │ ├── rag/
│ │ ├── graph/
│ │ └── chat/
│ ├── app.js
│ ├── server.js
│ └── package.json
│
├── .env.example
├── .gitignore
├── README.md
└── package.json

---

⚙️ Prerequisites

Install the following:

• Node.js 18+
• npm
• Git
• MongoDB / MongoDB Atlas
• Neo4j / Neo4j Aura
• Pinecone Account
• Google Gemini API Key
• Tavily API Key

---

📥 Installation & Setup

1. Clone Repository

git clone https://github.com/Vishal202-rgb/ResearchAI-MultiAgent.git
cd ResearchAI-MultiAgent

2. Install Backend Dependencies

cd server
npm install

3. Install Frontend Dependencies

Open another terminal:

cd client
npm install

---

🔐 Environment Variables

Create the following file:

server/.env

You can use .env.example as the template.

PORT=5000MONGODB_URI=your_mongodb_connection_stringJWT_SECRET=your_secure_jwt_secretJWT_EXPIRES_IN=7dCLIENT_URL=http://localhost:5173NODE_ENV=development# GeminiGEMINI_API_KEY=your_gemini_api_key# TavilySEARCH_API_KEY=your_tavily_api_keySEARCH_API_URL=https://api.tavily.com/search# PineconePINECONE_API_KEY=your_pinecone_api_keyPINECONE_INDEX=your_pinecone_index# Neo4jNEO4J_URI=your_neo4j_uriNEO4J_USER=neo4jNEO4J_PASSWORD=your_neo4j_password

⚠️ Never commit .env files or expose API keys publicly.

---

▶️ Run the Application

Start Backend

From the server directory:

npm run dev

Expected output:

MongoDB connected successfully
Neo4j connected
Server running on port 5000

Backend:

http://localhost:5000

Start Frontend

Open another terminal:

cd client
npm run dev

Frontend:

http://localhost:5173

Open the frontend URL in your browser.

---

🔐 Authentication Flow

User
│
▼
Register / Login
│
▼
JWT Token
│
▼
Protected API Routes
│
▼
Authenticated Workspace

Passwords are securely hashed using bcryptjs.

---

🌐 Web Search Flow

Research Question
│
▼
Researcher Agent
│
▼
Tavily Search API
│
▼
Real Web Sources
│
▼
Analysis + Fact Checking
│
▼
Final Research

---

🗄️ Database Responsibilities

Technology	Purpose
MongoDB	Users, workspaces, findings, reports and application data
Pinecone	Document embeddings and semantic retrieval
Neo4j	Entities and relationships for the knowledge graph
Gemini	AI reasoning, agents, synthesis and analysis
Tavily	Real-time web search and source discovery

---

🧪 Development Workflow

Run both services simultaneously:

Terminal 1
────────────────────────
cd server
npm run dev
Terminal 2
────────────────────────
cd client
npm run dev

Frontend:

http://localhost:5173

Backend:

http://localhost:5000

---

☁️ Deployment

ResearchAI can be deployed using Vercel.

Production Environment Variables

Add these variables in:

Vercel Dashboard
→ Project
→ Settings
→ Environment Variables

Required variables:

MONGODB_URI
JWT_SECRET
JWT_EXPIRES_IN
GEMINI_API_KEY
SEARCH_API_KEY
SEARCH_API_URL
PINECONE_API_KEY
PINECONE_INDEX
NEO4J_URI
NEO4J_USER
NEO4J_PASSWORD
CLIENT_URL
NODE_ENV

Deployment Checklist

1. Push code to GitHub
↓
2. Import repository into Vercel
↓
3. Configure environment variables
↓
4. Configure build settings
↓
5. Deploy
↓
6. Test authentication
↓
7. Test research workflow
↓
8. Test RAG
↓
9. Test Knowledge Graph
↓
10. Test PDF Export

Never store production secrets inside GitHub.

---

🔮 Future Scope

👥 Collaborative Research

Enable multiple users to collaborate inside the same research workspace.

Planned capabilities:

• Workspace sharing
• Viewer / Editor permissions
• Real-time collaboration
• Comments and annotations
• Activity history

Additional Improvements

• Research version history
• Citation credibility scoring
• Multi-model support
• Voice-based research assistant
• Browser extension
• Scheduled research agents
• Team research dashboards

---

💡 Why ResearchAI?

Traditional research requires manually:

Search
↓
Read
↓
Analyze
↓
Verify
↓
Compare
↓
Organize
↓
Write

ResearchAI brings these steps into one intelligent workspace:

Research Question
↓
Multi-Agent Research
↓
Real Web Sources + RAG Documents
↓
Knowledge Graph + Fact Checking
↓
Trace Evidence + Debate Findings
↓
Contradictions + What Changed?
↓
Comparison + Timeline
↓
Research Chat + Deep Dive
↓
Final Report + PDF Export

---

📌 Project Status

🟢 Production-Ready Portfolio Project

Currently implemented:

• Multi-Agent Research
• Real-Time Web Search
• PDF/TXT RAG
• Pinecone Vector Search
• Neo4j Knowledge Graph
• Trace Evidence & Debate Findings
• Contradiction Detection
• What Changed? (Research Evolution)
• Research Chat
• Deep Dive Research
• Workspace Comparison
• Research Timeline
• Global Search
• PDF Export
• JWT Authentication
• Demo Workspaces
• Responsive Premium UI
• Dark/Light Mode
• Vercel Deployment

---

👨‍💻 Author

Vishal Kumar

B.Tech Final Year Student
Generative AI & Full Stack Developer

---

⭐ Support

If you find ResearchAI useful, consider giving the repository a ⭐ on GitHub.

---

📜 License

This project is licensed under the MIT License.