|
Download README.md from RAGnTeX/ragntex: direct link, hf CLI and curl.
- Browser
- Download file 2.47 kB
-
https://huggingface.co/spaces/RAGnTeX/ragntex/resolve/main/README.md
- Command line
-
hf download hf://spaces/RAGnTeX/ragntex/README.md
-
curl -L -o README.md https://huggingface.co/spaces/RAGnTeX/ragntex/resolve/main/README.md
2.47 kB
| title: RAGnTeX | |
| emoji: 📄 | |
| colorFrom: indigo | |
| colorTo: pink | |
| sdk: docker | |
| pinned: false | |
| # 📚 From Text to Visuals: Auto-Generating LaTeX Beamer Presentations with GenAI | |
| In this project, we explore how generative AI can automate the creation of professional-looking presentation slides—directly from extensive collections of PDF documents. | |
| --- | |
| ## 🧠 Use Case | |
| Creating slide decks from dense documents (like whitepapers or scientific articles) is a time-consuming and cognitively heavy task. Our goal is to streamline this process using generative AI. | |
| We built an AI assistant that transforms document collections into LaTeX Beamer presentations—complete with structure, content, and visuals—based on a user-defined topic. | |
| --- | |
| ## 🔍 How It Works | |
| ### 💿 Database Creation | |
| We start by creating a vector database from a collection of PDF documents (e.g., arXiv papers). Each document is processed to extract text chunks, and for each chunk, we store: | |
| * **Embeddings** of the text from the file | |
| * **Associated metadata** (like images, their captions, how many images there are) | |
| ### 🗣️ User Prompt | |
| A user enters a natural language query like: | |
| > "I need a presentation about AI agents." | |
| ### 📄 Document Retrieval | |
| We use **Chroma** as our vector database to retrieve the most relevant PDF documents based on the query. | |
| ### 🧠 Content Understanding | |
| Using **Gemini 2.0 Flash** (`google-genai==1.7.0`), the assistant analyzes the retrieved content and extracts the most relevant ideas and insights. | |
| ### 🎞️ Slide Generation (LaTeX Beamer) | |
| The model outputs fully formatted compilable LaTeX code following a clear and consistent structure: | |
| - **Introduction**: Topic overview and motivation | |
| - **Main Content**: 2–4 slides, each presenting a core idea | |
| - **Summary**: Key takeaways | |
| It also incorporates relevant images extracted from the documents, organizing them into LaTeX Beamer-friendly layouts (e.g., *Core Idea 2* and *Core Idea 3* slide formats). | |
| ### Prerequisites | |
| This project requires the LaTeX toolchain, specifically `pdflatex`, to generate PDF presentations. | |
| Please install it before running the project: | |
| ### On Debian/Ubuntu: | |
| ```bash | |
| sudo apt-get update && sudo apt-get install -y texlive-latex-base texlive-fonts-recommended texlive-latex-extra | |
| ### Run docker: | |
| ```bash | |
| docker run \ | |
| --env-file .env \ | |
| -e IN_DOCKER=true \ | |
| -p 7860:7860 \ | |
| -v /path/to/your/output:/app/output \ | |
| ragntex | |