TypedDoc: Handwritten Notes to Digital Markdown & Formatted Documents

TypedDoc is an AI-powered web application that transcribes handwritten notes, university problem sets, lecture boards, and scanned PDFs into clean, publication-ready GitHub-flavored Markdown, LaTeX mathematics, DOCX, and formatted PDFs.

Powered by Google Gemini Multimodal AI with a 6-key parallel rotation pool and two-pass verification, TypedDoc intelligently reconstructs complex derivations, multi-step algebraic formulas, algorithms, truth tables, and technical ASCII diagrams with zero silent omissions.

---

✨ Features

• Gemini Vision OCR Pipeline: High-accuracy multimodal transcription powered by gemini-3.8-flash with automatic fallback tiers (gemini-3.5-flash, gemini-3.5-flash-lite).
• Parallel Multi-Key KeyPool: Dynamically distributes multi-page PDF processing across up to 6 Gemini API keys with per-key RPM pacing and independent quota tracking.
• Authoritative Document Generation: Produces identical structured outputs for live Markdown preview, formatted Word documents (.docx), and styled academic PDFs (.pdf).
• Structured OCR AST & Two-Pass Verification: Validates LaTeX mathematics balance ($$), Markdown table column consistency, code indentation, and diagram bounds with confidence scoring.
• Content-Addressed Safe Caching: SHA-256 caching with atomic temp-file renaming ensures instant resumption and zero duplicated API spend if interrupted.
• Browser Fallback Mode: Automatically falls back to in-browser client-side OCR (Tesseract.js + PDF.js) if network or server connection is interrupted.
• Interactive Document Editor: Split-view and full-preview modes, debounced autosave, snapshot version control, live stats, and keyboard shortcuts (Ctrl+S).
• Cloud Storage: Secure Firebase Authentication and Firestore document synchronization.

---

🛠️ Tech Stack

Frontend

• Framework: React 19 with Vite 6
• Styling: Tailwind CSS & Motion animations
• Math Rendering: KaTeX
• Persistence: Firebase SDK (Auth & Firestore)
• Icons: Lucide React

Backend

• Server: Node.js & Express 5
• Multimodal AI: Google Gemini API (@google/genai)
• PDF Generation: pdf-lib (pure JS, low-memory footprint <30MB)
• DOCX Generation: docx library
• Security & Reliability: Helmet, Compression, and user-aware rate limiting

---

🚀 Quick Start

Prerequisites

• Node.js ≥ 20.0.0
• npm ≥ 9.0.0
• One or more Google Gemini API keys (Get your free API keys from Google AI Studio)
• Firebase project credentials

1. Installation

git clone https://github.com/sandeshpawar10/HandWritten-PDF-to-Typed-PDF.git
cd typed-document-v2
npm install

2. Configure Environment Variables

Copy .env.example to .env:

cp .env.example .env

Configure your Gemini API key(s) in .env:

# Single key or comma-separated list:
GEMINI_API_KEY=your_gemini_key_1,your_gemini_key_2

# Or named individual keys (supports up to 6 or more):
GEMINI_KEY_01=your_key_1
GEMINI_KEY_02=your_key_2

# Active model tiers:
GEMINI_MODEL_PRIMARY=gemini-3.8-flash
GEMINI_MODEL_FAST=gemini-3.5-flash
GEMINI_MODEL_VERIFY=gemini-3.8-flash
GEMINI_MODEL_FALLBACK=gemini-3.5-flash-lite

# Concurrency (matches active keys):
OCR_CONCURRENCY=6

PORT=3001
NODE_ENV=development

3. Run Development Server

Run frontend and backend concurrently:

npm run dev:all

• Frontend: http://localhost:3000
• Backend API: http://localhost:3001

---

🧪 Testing

Run the automated test suite:

npm test

Run test suite in watch mode:

npm run test:watch

Run linter:

npm run lint

Build production bundle:

npm run build

---

📄 License

MIT License.