Book Translate

A browser-based tool for translating PDF books and markdown documents to any language using OpenAI's GPT models. Upload a PDF, clean up OCR errors, and get professional translations—all from your browser.

Overview

Book Translate handles the complete document translation pipeline:

1. Convert - Transform PDFs into editable markdown
2. Clean Up - Fix OCR errors and PDF conversion artifacts
3. Translate - Get natural, context-aware translations in any language

All processing happens client-side in your browser. Your documents and API key never leave your machine (except for API calls directly to OpenAI).

Quick Start

1. Open the App

Visit science.github.io/ai-translator to launch the app in your browser.

2. Configure Your API Key

Navigate to Settings and enter your OpenAI API key. Need help? Click "How to get an API key" for step-by-step instructions including:

• Creating an OpenAI account
• Setting up billing (required for API access)
• Generating and saving your key

3. Translate Your First Document

Option A: One-Step Translation (Recommended)

1. Go to One Step Translation in the sidebar
2. Upload a PDF or markdown file
3. Enter your target language (e.g., "Japanese", "formal German")
4. Click Start and wait for processing
5. Download your translated document

Option B: Step-by-Step

1. Upload your document on the Home page
2. Convert it on the Convert PDF page
3. Clean it up on the Cleanup page
4. Translate it on the Translate page
5. Download from My Documents

Features

One Step Translation

The flagship feature—a unified pipeline that handles everything:

• Upload PDF or markdown → Automatic conversion (if needed), cleanup, and translation
• Configure cleanup and translation settings independently
• Track progress through each phase with visual indicators
• Download as Markdown or Word document

Outputs:

• Converted Markdown (raw PDF extraction, PDF input only)
• Cleaned Markdown (OCR errors fixed)
• Translated (target language only)
• Bilingual (alternating original/translated sections)

PDF to Markdown Conversion

Extract text from PDF files while preserving structure:

• Headers, paragraphs, and formatting maintained
• Fast local processing (no API calls required)
• Works with text-based and scanned PDFs

Document Cleanup (Rectification)

Fix common OCR and PDF conversion errors:

• Missing letters: ontents → Contents
• Broken words: Ww hile → While
• Page numbers and footer markers removed
• Paragraph flow restored across page breaks
• PDF gibberish text removed

Translation

Natural, publication-quality translations:

• Target language flexibility - Specify language and tone (e.g., "formal Japanese", "conversational Spanish")
• Language history - Quick access to recently used languages
• Context-aware translation - Optional feature that provides surrounding text context for consistent tone
• Dual output - Get target-language-only and bilingual versions

Document Management

Full document library with organization features:

• Phase filtering - View by stage: Uploaded, Converted, Cleaned, Translated
• Search - Find documents by name
• Preview - View raw markdown or rendered HTML
• Export - Download as original format or Word document
• Storage tracking - Monitor browser storage usage

Settings

Configure defaults and manage your workspace:

• API key configuration with connection testing
• Default model, chunk size, and reasoning effort
• Context-aware translation toggle
• Storage management with one-click cleanup

Supported Models

Model	Speed	Quality	Best For
gpt-5.2	Slower	Highest	Final translations
gpt-5-mini	Medium	High	General use (default)
gpt-4.1	Fast	Good	Quick drafts
gpt-4.1-mini	Fastest	Good	Large documents

GPT-5 models support configurable reasoning effort (Low/Medium/High) for balancing speed vs. quality.

User Guide

Choosing a Workflow

Use One Step Translation when:

• You have a PDF or markdown file that needs translation
• You want the simplest experience
• You need all output formats

Use individual pages when:

• You only need PDF conversion (no translation)
• You want to review/edit between steps
• You need more control over each phase

Target Language Tips

The target language field accepts natural descriptions:

• Japanese - Standard translation
• Formal Japanese - Business/academic tone
• Conversational Brazilian Portuguese - Casual, regional style
• Simplified Chinese for technical readers - Domain-specific adaptation

Your recent languages are saved for quick reuse.

Document Phases

Documents progress through phases, visible in My Documents:

Phase	Description
Uploaded	Original file, unprocessed
Converted	PDF extracted to markdown
Cleaned	OCR errors and artifacts fixed
Translated	Final translation (with variant tags)

Storage and Privacy

• Local storage only - Documents stored in your browser's IndexedDB
• API key security - Stored locally in localStorage, never transmitted except to OpenAI
• No server - All processing is client-side
• Clear anytime - Delete all documents from Settings

CLI Tool

For automation or batch processing, a command-line interface is also available:

# Convert PDF to markdown
node src/index.js book.pdf --pdf-to-md --output-dir converted/

# Clean up OCR errors
node src/index.js converted/book.md --rectify --output-dir cleaned/

# Translate to Japanese
node src/index.js cleaned/book-rectified.md --output-dir translated/

See CLI Documentation below for full options.

Development

Prerequisites

• Node.js 20 LTS or later
• OpenAI API key with billing enabled

Running Your Own Copy

To host the web app locally or deploy your own instance:

cd web-app
npm install
npm run dev          # Start development server at http://localhost:5173
npm run build        # Build for production (outputs to build/)
npm run preview      # Preview production build locally

Web App Development

cd web-app
npm run dev          # Start development server
npm run build        # Build for production
npm run check        # TypeScript type checking
npm run test:unit    # Run Vitest unit tests
npm run test:e2e     # Run Playwright browser tests
npm run test:e2e:ui  # Playwright with interactive UI

CLI Development

npm install
npm test                              # Run all Jest tests
npm test -- --testPathPattern=chunker # Run specific test file

Tech Stack

Web App:

• SvelteKit 2 / Svelte 5
• Tailwind CSS 4
• TypeScript
• IndexedDB (via idb)
• Vitest + Playwright

CLI:

• Node.js ES Modules
• OpenAI SDK
• Jest

Project Structure

book-translate/
├── web-app/                    # Browser application
│   ├── src/
│   │   ├── routes/            # SvelteKit pages
│   │   │   ├── +page.svelte           # Home/Upload
│   │   │   ├── workflow/              # One Step Translation
│   │   │   ├── convert/               # PDF to Markdown
│   │   │   ├── cleanup/               # Document rectification
│   │   │   ├── translate/             # Translation
│   │   │   ├── documents/             # Document library
│   │   │   └── settings/              # Configuration
│   │   └── lib/
│   │       ├── stores/        # State management
│   │       ├── services/      # API clients, document operations
│   │       └── components/    # Reusable UI components
│   └── tests/                 # E2E tests
├── src/                       # CLI tool
│   ├── index.js              # CLI entry point
│   ├── chunker.js            # Document chunking
│   ├── translator.js         # OpenAI translation
│   ├── rectifier.js          # OCR error correction
│   └── pdfConverter.js       # PDF extraction
└── test/                      # CLI tests

CLI Reference

node src/index.js <input-file> [options]

Modes (mutually exclusive):

• --pdf-to-md - Convert PDF to markdown
• --rectify - Clean up OCR errors (English to English)
• (default) - Translate to Japanese

Options:

• --output-dir <path> - Output directory (default: output/)
• --chunk-size <n> - Characters per chunk (default: 4000)
• --model <name> - OpenAI model (default: gpt-5-mini)
• --reasoning-effort <level> - GPT-5 reasoning: low, medium, high

Examples:

# Full pipeline
node src/index.js book.pdf --pdf-to-md --output-dir step1/
node src/index.js step1/book.md --rectify --output-dir step2/
node src/index.js step2/book-rectified.md --output-dir final/

# Quick translation with faster model
node src/index.js clean-book.md --model gpt-4o --chunk-size 3000

Environment Setup

Create .env from the template:

cp .env.example .env

Add your OpenAI API key:

OPENAI_API_KEY=sk-...

Troubleshooting

"Insufficient quota" error

• Add credits to your OpenAI account at platform.openai.com/settings/organization/billing

"Rate limit exceeded" error

• The app automatically retries with backoff
• New accounts have lower limits that increase over time
• Try reducing chunk size or using a faster model

API key not working

• Ensure billing is set up (required even with free credits)
• Check the key hasn't been revoked
• Verify no extra spaces when pasting

Poor PDF conversion quality

• Scanned PDFs produce more OCR errors—always run Cleanup
• Some complex layouts may not convert perfectly

Dependencies

Runtime:

• openai - OpenAI API client
• @opendocsg/pdf2md - PDF to markdown conversion
• idb - IndexedDB wrapper
• marked - Markdown rendering
• @mohtasham/md-to-docx - Word document export

Development:

• jest / vitest - Testing frameworks
• playwright - Browser testing
• svelte / sveltekit - Web framework
• tailwindcss - Styling