Document Localization Studio

This is a submission for the GitHub Copilot CLI Challenge

What I Built

I built Document Localization Studio — a terminal-first + UI-powered app that localizes documents beyond basic translation.

Instead of treating localization as “just translate text,” this project tackles the real-world complexity teams hit in enterprise docs:

• 🌐 Language + terminology adaptation (custom glossary + reusable term memory)


• 🗓️ Date/time + timezone conversion (e.g., America/New_York → Europe/Berlin)


• 💱 Currency + FX conversion (USD → EUR/JPY/BRL/… with locale defaults you can edit)


• 📏 Unit conversion (mi→km, lb→kg, °F→°C)


• 📬 Address/phone/postal tweaks (locale labels + phone formatting)


• 🧾 Tax label adaptation (Sales Tax → VAT/GST-style labels)


• 🔒 Legal clause lock/protection ([[LOCK]]...[[/LOCK]] + auto-protect legal-ish sentences)


• ✅ Structure-aware QA (placeholders preserved, length-change warnings, cross-ref/TOC flags, workflow gating)

Supported file types 🧩

• .txt


• .docx


• 
  .pdf
  
  
  
  
  • 
    layout-preserving mode for editable PDFs (when available)
  
  
  
  
  
  



• screenshots/images: .png, .jpg, .jpeg via OCR

Supported locales 🗺️

de_de, es_es, fr_fr, it_it, ja_jp, ko_kr, pt_br, zh_cn, zh_tw

Run locally 🧪

cd "/Users/swatigoyal/Documents/New project/document_localizer_challenge"

python3 -m venv .venv

source .venv/bin/activate

pip install -r requirements.txt

streamlit run app.py

CLI example 🧰

python -m localizer.cli input.pdf output.pdf \ --locale de_de \ --source-timezone America/New_York \ --tone legal

Demo

• 
  Repo: https://github.com/swatigoyal911/document_localizer_challenge
  


• 
  Live demo: https://youtu.be/yX4bfBfwlMk
  

Walkthrough idea (video/screenshots) 🎬

• Upload a real invoice/contract PDF (or a DOCX).


• Pick a target locale (ex: de_de) and watch the default FX rate auto-load (editable).


• Toggle components (units, tax labels, legal lock, term memory).


• Run localization.


• Show the outputs:
  
  
  
  
  • 📊 Before/After scorecards
  
  
  • 🔎 Side-by-side visual diff
  
  
  • 🌡️ Layout risk heatmap
  
  
  • 🧾 QA report (JSON)
  
  
  
  
  
  



• Download the localized file + QA report.

Stack / Libraries 🧱🐍

Built with a “free stack”:

• 
  streamlit (UI dashboard)


• 
  python-docx (DOCX read/write)


• 
  pypdf (PDF text extraction)


• 
  pymupdf (PyMuPDF — layout-preserving PDF localization mode)


• 
  reportlab (PDF re-render fallback when layout mode isn’t available)


• 
  pillow + pytesseract (OCR pipeline for screenshots/images)

OCR note: screenshot localization requires a local Tesseract binary in addition to pytesseract (ex: macOS brew install tesseract).

My Experience with GitHub Copilot CLI 🤝⚡

I used GitHub Copilot CLI as a coding partner directly in the terminal to:

• 🏗️ scaffold modules quickly (pipeline, PDF/DOCX/image IO, CLI wiring)


• 🧠 iterate on regex-heavy transformations (dates, currency, units, placeholders)


• 🧩 design locale profiles/defaults and keep the logic consistent


• 🎛️ wire Streamlit controls to the backend config without breaking flow


• 🧪 add QA heuristics + sensible fallback paths for PDFs/OCR


• 🧹 speed up refactors while keeping the project clean and extensible

The biggest win: fast iteration on non-trivial logic (PDF handling + transformation rules + feature toggles) without leaving the terminal.

What’s Next / Improvements 🚀🤖

This is a strong prototype — and there’s a lot we can level-up with AI integration later:

• 🧠 LLM-backed translation (while keeping deterministic transforms + locks)


• 📚 smarter terminology alignment (context-aware term choice + consistency scoring)


• 🧾 stronger compliance checks (policy packs per industry/locale)


• 🧩 plug-in architecture for new transforms + QA rules


• 🖼️ better OCR layout reconstruction (tables, columns, headers/footers)

If you’ve worked on localization, I’d love your feedback: what transformations or QA checks would you trust most in production?