Desktop App Pairs Your PDFs with a Grumpy AI Professor

What it does This is a Python desktop application built for grinding through academic papers. Drop in a PDF and it extracts the text via MinerU, translates it between Chinese and English, builds a FAISS vector index, and opens a split-screen view: paper on the left, AI chat on the right. You can type questions or talk to it via real-time voice input, and the app answers using RAG retrieval against the document while reading responses aloud through MiniMax TTS.

The interesting bit The hook is the persona engine. The default assistant is a 暴躁 (grumpy) professor, but you can hot-swap character prompts by dropping text files into a prompt folder—the repo ships with examples named keli and leidian. The system even maps detected emotion to TTS voice settings, so the professor’s tone shifts with the conversation.

Key highlights

• End-to-end pipeline from PDF to searchable vector store, powered by MinerU and FAISS
• Bilingual reading mode with a UI toggle between Chinese and English
• Real-time voice loop using RealtimeSTT for input and MiniMax streaming TTS for output
• RAG-backed Q&A that grounds responses in the imported paper
• Swappable AI personas via plain text prompt files
• Requires a CUDA-capable GPU with 6GB+ VRAM

Caveats

• Strictly for structured academic papers; non-standard PDFs may crash or fail to parse
• Voice chat without headphones creates a feedback loop where the AI’s voice is re-captured as user input
• Changing the professor’s persona or voice requires manually editing Python source files and prompt paths

Verdict Try it if you want a localized, bilingual paper assistant with actual personality and don’t mind editing code to customize it. Look elsewhere if you need broad PDF compatibility, a fully offline stack, or a polished settings UI.