S-Translate — س-ترجم

A document translator that keeps your formatting intact. Upload a Word, PDF, PowerPoint, Excel, or text file, pick a target language, and get back the same document — same layout, tables, and right-to-left direction — translated.

Challenge

Generic translators flatten documents: they lose tables, styles, images, and especially Arabic RTL layout. Re-formatting a translated document by hand often takes longer than the translation itself.

Solution

• Format-preserving pipeline: parses each format with the right tooling (python-docx, python-pptx, openpyxl, PyMuPDF) and re-emits the original structure with translated text in place.
• Model-agnostic AI: translation runs through LiteLLM, so any provider works (OpenAI, Anthropic, Gemini, DeepSeek, or a fully-local Ollama) — users can bring their own API key or use a shared one.
• Arabic-aware: handles bidirectional text, reshaping, and embedded fonts; falls back to OCR (Tesseract) for scanned PDFs.
• Self-hostable: stateless, file-based jobs with a one-command Docker setup.

Technical Implementation

A FastAPI (Python) backend serving a React + Vite + Tailwind front end, with an async job queue and ~64 offline tests against a stubbed model. Ships as a multi-stage Docker image with an optional local-LLM (Ollama) profile.