LocalScholar-Flow: Private, Offline Academic Paper Translation with Perfect Formatting (Docker + MinerU + Hunyuan)

[littleBu0210]

Hi everyone! 👋

I built LocalScholar-Flow, an open-source pipeline designed to solve the biggest headache in academic research: Translating papers while keeping formulas, images, and layout intact, without uploading data to the cloud.

We all know standard translators break mathematical formulas and destroy document structure. I combined the best open-source tools into a single, easy-to-deploy Docker workflow.

👀 See the Results:

1. PDF to Markdown Parsing (Powered by MinerU)
It doesn't just extract text; it understands the layout.
(Check out the comparison images below/attached: Note how the complex layout and headers are preserved)

2. Context-Aware Translation (Powered by Tencent Hunyuan)
Using a local LLM specifically tuned for translation, it keeps the academic tone and preserves LaTeX formulas.
(See the translation output: The math formulas remain perfect inside the translated text)

✨ Why use this?

• 🔒 100% Privacy: Everything runs locally on your GPU. No data leaves your machine.
• 🐳 One-Click Docker Setup: No dependency hell. Just run python scripts/download_models.py and docker compose up -d.
• ⚡ Batch Processing: State management via MongoDB allows you to process folders of PDFs and resume if interrupted.
• 🌍 Cross-Platform: Optimized scripts for Windows, Linux, and macOS.

🛠️ Tech Stack:

• Parser: MinerU (OpenDataLab)
• Translator: HY-MT (Hunyuan) running on vLLM
• Backend: Python + MongoDB

🔗 GitHub Repository:
👉 https://github.com/littleBu0210/LocalScholar-Flow

I’d love to hear your feedback! If this helps your research workflow, please consider giving it a ⭐!