Deep Research Citation Verifier (Athena DeepCite)

A Chrome extension that helps users verify factual claims on webpages and PDFs. It combines AI-powered claim detection with Exa's neural search technology to provide source verification and confidence scores for factual statements.

Features

• Detects factual claims in web pages and PDFs on demand
• Uses AI-powered claim extraction with confidence scoring
• Highlights claims with a subtle blue background
• Shows multiple sources on hover with confidence scores
• Provides direct links to original source documents
• Supports dark mode through CSS variables
• Works with both web pages and PDF documents

Installation

1. Clone the repository

git clone [repository-url]
cd athena-deepcite

2. Install dependencies

npm install

3. Build the extension

npm run build

4. Load in Chrome

• Open Chrome and go to chrome://extensions/
• Enable "Developer mode" in the top right
• Click "Load unpacked"
• Select the extension directory containing manifest.json

Project Structure

The project follows a modular architecture with clear separation of concerns:

├── src/                   # Source TypeScript code
│   ├── extractors/        # Claim extraction strategies
│   │   ├── contentExtractor.ts # Rule-based extraction
│   │   └── llmExtractor.ts     # AI-powered extraction
│   ├── handlers/          # Document-specific processing
│   │   ├── pdfHandler.ts      # PDF document handling
│   │   └── webPageHandler.ts  # Web UI components
│   ├── utils/             # Shared utilities
│   │   └── settingsManager.ts # Storage and settings mgmt
│   ├── background.ts      # Background service worker
│   ├── contentScript.ts   # Main content script
│   ├── options.ts         # Options page functionality
│   ├── popup.ts           # Popup UI functionality
│   └── types.ts           # Shared TypeScript interfaces
├── dist/                  # Compiled JavaScript
├── lib/                   # Third-party libraries (pdf.js)
├── styles/                # CSS styles with variables
│   └── pdf-overlay.css    # Styling for web and PDF overlays
├── options.html           # Settings page UI
├── popup.html             # Popup UI
└── manifest.json          # Extension config and permissions

Usage

1. Visit any webpage with factual content (e.g., news articles, Wikipedia, research blogs)

2. Click the DeepCite button in the toolbar to open the popup menu

3. Select "Analyze Current Page" to detect and highlight claims.

4. The extension will:

   • Find factual claims in the text using either AI or rule-based analysis
   • Highlight them with a subtle blue background
   • Connect to Exa's API to verify sources
   • Show a sidebar with all detected claims

5. Interact with highlights:

   • Hover over any highlighted text
   • See source information and confidence scores in a tooltip
   • Click "View source" to open the original reference

Development

Building and Watching

# Build once
npm run build

# Watch for changes during development
npm run dev

# Run linting
npm run lint

# Clean build output
npm run clean

API Keys Configuration

The extension stores API keys securely in Chrome's storage API. Keys can be configured in the options page:

1. Exa API: Used for retrieving source information for claims

   • Create an account at https://exa.ai to get your API key
   • Enter your key in the options page

2. OpenAI API (Optional): Used for improved claim detection and confidence scoring

   • Create an account at https://platform.openai.com to get your API key
   • Enter your key in the options page to enable AI-powered extraction
   • Without this key, the extension will use rule-based extraction instead

TypeScript Path Aliases

The project uses TypeScript path aliases for cleaner imports:

• @extractors/* - Points to src/extractors/
• @handlers/* - Points to src/handlers/
• @utils/* - Points to src/utils/
• @types - Points to src/types.ts

Key Components

• Claim Detection: Uses heuristics or LLM to find factual statements
• Source Verification: Calls Exa API to find supporting sources
• UI: Highlights + tooltips + sidebar showing source info
• Settings Management: Secure storage and configuration

Testing on Different Sites

Works best on:

• Wikipedia articles
• News sites
• Research blogs
• Technical documentation
• PDF documents (research papers, reports)

Next Steps

• Add manual analysis button to respect user privacy
• Support multiple claim highlights
• Improve UI with modern design elements
• Add confidence scores from LLM analysis
• Refactor code for better maintainability
• Improve cross-site compatibility for highlighting
• Add error handling for API failures
• Add browser dark mode support
• Package for Chrome Web Store

Development Documentation

For developers working on this extension, please see the following documentation:

• Current Status
• Testing Guide
• Refactoring Plan
• Implementation Plan