IntelliRead - PDF Intelligence Platform

Transform enterprise PDFs into searchable, structured knowledge using AI-powered analysis.

Overview

IntelliRead is a local-first, browser-only application that converts PDFs into searchable structured knowledge. All processing happens client-side using IndexedDB for storage - no backend servers, no external databases.

The Problem It Solves

Organizations manage thousands of PDFs: manuals, reports, policies, research papers. Finding specific information across these documents is time-consuming because:

• PDF content is unstructured
• Search tools only find exact text matches
• Images, charts, and scanned pages are invisible to search
• No way to ask questions and get contextual answers

IntelliRead transforms PDFs into queryable knowledge bases where you can search semantically and ask natural language questions.

Supported PDF Types

Type	Description	Processing Method
Text-based PDFs	Standard PDFs with selectable text	Direct text extraction via PDF.js
Mixed PDFs	Documents combining text and images	Hybrid extraction with AI image analysis
Image-only PDFs	Scanned documents, photo PDFs	Full page rendering + Gemini AI descriptions
Chart-heavy PDFs	Data visualizations and diagrams	AI-powered chart interpretation

Core Features

Intelligent Document Processing

• Structure Preservation: Maintains headings, sections, chapters, and page numbers
• Smart Chunking: Splits content at natural boundaries (sentences, paragraphs, sections) - never mid-word
• Table Extraction: Detects and extracts tables with row/column integrity
• Image Analysis: AI-generated descriptions make visual content searchable

AI-Powered Search & Q&A

• Semantic Search: Find content by meaning, not just keywords
• Multi-Provider AI: Choose between Groq, Perplexity, Anthropic, or Google Gemini
• Cited Answers: AI responses include page references for verification
• Document-Scoped: Answers are grounded in your uploaded documents

Privacy-First Architecture

• 100% Client-Side: All processing happens in your browser
• IndexedDB Storage: Documents persist locally, never uploaded to servers
• API Calls Only When Needed: AI providers are called only for:
  • Generating image descriptions (Gemini)
  • Answering questions (Groq/Perplexity/Anthropic)

Technology Stack

Layer	Technology
Frontend	React 18 + TypeScript + Vite
UI Components	shadcn/ui + Tailwind CSS
PDF Processing	PDF.js (client-side)
Storage	IndexedDB via idb library
AI Providers	Groq, Perplexity, Anthropic, Google Gemini

Getting Started

Prerequisites

• Node.js 18+ and npm
• Required API keys (these three must be configured):
  • Google Gemini (for image analysis)
  • Groq (for fast Q&A)
  • Perplexity (for web-grounded answers)
• Optional API key:
  • Anthropic Claude (for complex reasoning)

Installation

# Clone the repository.
git clone https://github.com/Illusory-warden/intelliread.git
cd intelliread

# Install dependencies
npm install

# Start development server
npm run dev

Configuration

1. Open IntelliRead in your browser
2. Click API Settings in the header
3. Enter API keys for the required providers:
   • Gemini (Required): For image analysis (Get key)
   • Groq (Required): For fast LLM responses (Get key)
   • Perplexity (Required): For web-grounded answers (Get key)
   • Anthropic (Optional): For Claude models (Get key)
4. Test each connection before saving
5. PDF upload is disabled until all required keys are configured

Usage

1. Upload a PDF: Drag and drop or click to select
2. Wait for Processing: Watch the progress indicator
3. Explore Content: Browse sections, tables, and images
4. Ask Questions: Use the chat interface to query your document
5. Export: Download chat history as PDF

System Flow

PDF Upload → Page Analysis → Content Extraction → Chunking → Indexing → Search Ready
                 ↓
         Image Detection
                 ↓
         AI Description (Gemini)
                 ↓
         Content Normalization

For detailed architecture, see docs/ARCHITECTURE.md.

Limitations

Limitation	Description
Image Interpretation	AI descriptions are interpretive, not exact transcriptions
Complex Charts	Multi-layered visualizations may not be fully captured
OCR Accuracy	Depends on scan quality; handwritten text may not process well
Offline AI	Image analysis and Q&A require internet connection
Browser Storage	IndexedDB has browser-specific storage limits

For complete limitations, see docs/LIMITATIONS.md.

Project Structure

src/
├── components/           # React UI components
│   ├── ui/              # shadcn/ui base components
│   ├── Header.tsx       # App header with navigation
│   ├── PDFViewer.tsx    # Document viewer with tabs
│   ├── ChatInterface.tsx # Q&A chat panel
│   └── ...
├── hooks/               # Custom React hooks
│   ├── useAPIKeys.ts    # API key management
│   ├── useChat.ts       # Chat state and API calls
│   └── useDocuments.ts  # Document CRUD operations
├── lib/
│   ├── apiClient.ts     # AI provider integrations
│   ├── db.ts            # IndexedDB operations
│   ├── imageExtractor.ts # PDF image detection
│   ├── pdfProcessor.ts  # PDF parsing and extraction
│   ├── textChunker.ts   # Content chunking logic
│   └── vectorSearch.ts  # Embedding and similarity
├── pages/
│   ├── Index.tsx        # Main application page
│   └── HowItWorks.tsx   # Processing explanation page
└── types/               # TypeScript definitions

Documentation

• Architecture Overview
• Image Processing Pipeline
• Processing Flow
• Indexing & Search
• Known Limitations
• Project Structure

License

MIT

Contributing

Contributions are welcome. Please open an issue first to discuss proposed changes.

Author

Illusory-warden