AI Knowledge Base - RAG Application

A powerful RAG (Retrieval Augmented Generation) based AI knowledge base built with Next.js, Gemini AI, and Pinecone vector database.

🚀 Features

• PDF Upload: Drag-and-drop interface for uploading PDF documents
• Smart Chunking: Automatically splits documents into digestible chunks with overlap
• AI Embeddings: Uses Google's Gemini model to generate vector embeddings
• Vector Storage: Stores embeddings in Pinecone for fast semantic search
• Chat Interface: Ask questions and get answers based on your documents
• Source Citations: See which parts of your documents were used to answer questions
• Beautiful UI: Premium gradient design with smooth animations

🛠️ Tech Stack

• Framework: Next.js 16 with App Router
• Styling: Tailwind CSS
• LLM: Google Gemini 3 API (gemini-2.0-flash-exp)
• Embeddings: Gemini text-embedding-004 model
• Vector Database: Pinecone
• PDF Processing: LangChain PDF Loader
• TypeScript: Full type safety

📋 Prerequisites

• Node.js 18+ and npm
• Pinecone account and API key
• Google AI (Gemini) API key

🔧 Setup

1. Clone and install dependencies:

npm install

2. Set up environment variables:

Create a .env.local file in the root directory:

# Pinecone API Configuration
PINECONE_API_KEY=your_pinecone_api_key_here
PINECONE_INDEX_NAME=knowledge-base

# Google Gemini API Configuration
GOOGLE_API_KEY=your_google_api_key_here

3. Run the development server:

npm run dev

4. Open your browser: Navigate to http://localhost:3000

🎯 How It Works

RAG Architecture

1. Document Ingestion:

   • User uploads a PDF
   • PDF is parsed and text is extracted
   • Text is cleaned and split into ~1000 character chunks with 200 character overlap

2. Embedding Generation:

   • Each chunk is converted to a 768-dimensional vector using Gemini
   • Vectors are stored in Pinecone with metadata (filename, chunk index, text)

3. Query Processing:

   • User asks a question
   • Question is converted to a vector embedding
   • Pinecone finds the top 5 most similar document chunks

4. Answer Generation:

   • Retrieved chunks are used as context
   • Gemini generates an answer based only on the context
   • Sources are shown with chunk references

📁 Project Structure

ai-knowledge-base/
├── src/
│   ├── app/
│   │   ├── api/
│   │   │   ├── upload/route.ts    # PDF upload & processing
│   │   │   └── chat/route.ts      # Chat/query endpoint
│   │   ├── layout.tsx
│   │   ├── page.tsx               # Main UI
│   │   └── globals.css
│   ├── components/
│   │   ├── FileUpload.tsx         # Drag-drop upload UI
│   │   └── ChatInterface.tsx      # Chat interface
│   ├── lib/
│   │   ├── pinecone.ts            # Pinecone client
│   │   ├── gemini.ts              # Gemini AI utilities
│   │   └── textUtils.ts           # Text chunking
│   └── types/
│       └── index.ts               # TypeScript types
├── .env.local
└── package.json

🚀 Deployment to Vercel

1. Push your code to GitHub

2. Connect to Vercel:

   • Go to vercel.com
   • Import your GitHub repository
   • Vercel will auto-detect Next.js

3. Add environment variables: In Vercel dashboard → Settings → Environment Variables, add:

   • PINECONE_API_KEY
   • PINECONE_INDEX_NAME
   • GOOGLE_API_KEY

4. Deploy:

   • Vercel will automatically deploy
   • Your app will be live at https://your-app.vercel.app

🔑 API Keys Setup

Pinecone

1. Sign up at pinecone.io
2. Create a new serverless index (dimension: 768, metric: cosine)
3. Copy your API key from the dashboard

Google Gemini

1. Go to Google AI Studio
2. Create an API key
3. Enable the Gemini API

📝 Usage

1. Upload a PDF:

   • Drag and drop a PDF or click to select
   • Wait for processing (you'll see a success message)

2. Ask Questions:

   • Type your question in the chat input
   • Get AI-generated answers based on your document
   • View source citations showing which parts were used

3. Upload More Documents:

   • Click "Upload another document" to add more PDFs
   • All documents are searchable in the same knowledge base

🎨 Customization

• Chunk Size: Modify chunkText() parameters in src/lib/textUtils.ts
• Number of Sources: Change topK in src/app/api/chat/route.ts
• UI Colors: Update Tailwind classes in components
• Models: Change model names in src/lib/gemini.ts

🐛 Troubleshooting

• "Cannot find module" errors: Run npm install again
• Pinecone errors: Ensure index dimension is 768 and metric is cosine
• API errors: Check your environment variables are set correctly
• Build errors: Try npm run build to see detailed error messages

📄 License

MIT

🤝 Contributing

Contributions welcome! Please open an issue or PR.

---

Built with ❤️ using Next.js, Gemini AI, and Pinecone