AutoDeckAI: 🌿 Eco-centric Slide Generator

Bridge the Gap Between Ecological Research and Impactful Communication
EcoHack-2025 - AI & LLM Hackathon for Applications in Evidence-based Ecological Research & Practice

![Demo Preview]

🚨 Problem Statement

Ecologists face critical challenges in research communication:

• Time-Consuming Workflows: 40+ hours spent monthly converting complex data into presentations
• Audience Mismatch: One-size-fits-all slides fail researchers, practitioners, and funders alike
• Visualization Bottlenecks: Manual extraction of 85%+ figures/tables from PDF manuscripts
• Draft Limitations: No tools adapt to preliminary abstracts/supplementary materials

🎯 Solution Highlights

AI-Powered Research Translation

• Audience-Specific Adaptation
  Tailor content depth for:

  • Researchers (Technical details)
  • Practitioners (Actionable insights)
  • Funding Bodies (Impact metrics)

• Intelligent Content Extraction

  graph LR
      A[PDF/Abstract] --> B[PyMuPDF Extraction]
      B --> C{Component Type?}
      C -->|Figure| D[BLIP Captioning]
      C -->|Table| E[GPT-4 Analysis]
      C -->|Text| F[FAISS Vectorization]
  

  Loading

• Draft-Friendly Processing
  Generate complete presentations from:

  • Partial manuscripts
  • Conference abstracts
  • Supplementary datasets

🌟 Key Features

Category	Capabilities
Input Flexibility	PDFs • DOCX • PPTX • TXT • Raw abstracts
AI Core	GPT-4o • Mixtral-8x7B • BLIP models • FAISS semantic search
Ecology Focus	Term validation • Sustainability metrics • Domain-specific visual themes
Output Quality	APA/MLA citations • Automated captions • 4K-ready vector graphics

🚀 Getting Started

Installation

git clone https://github.com/javadr/AutoDeckAI/
cd AutoDeckAI
python -m venv ecoenv && source ecoenv/bin/activate
pip install -r requirements.txt

Essential Dependencies:

pymupdf>=1.22.3  # PDF extraction
langchain>=0.1.5  # AI pipelines
faiss-cpu>=1.7.4  # Vector search
python-pptx>=0.6.21  # PPTX generation

Configuration

export OPENAI_API_KEY="sk-your-key"  # For GPT-4 integration
export HF_TOKEN="hf-your-token"     # Optional for open-source models

📊 Workflow Architecture

graph TD
        A[User Input] --> B{Input Type?}
        B -->|PDF/DOCX| C[PyMuPDF Extraction]
        B -->|Abstract| D[Direct Processing]
        C --> E[Component Classification]
        E --> F[Figure/Table Detection]
        F --> G[Caption Generation]
        G --> H[FAISS Vector Store]
        D --> H
        H --> I[Audience-Specific Prompting]
        I --> J[Slide Generation]
        J --> K[PPTX Assembly]
        K --> L[Output Validation]
        L --> M[Download Ready]

Loading

🔮 Future Roadmap

• Classroom Mode: Generate lecture decks from multiple papers
• Modular Decks: Create reusable slide components
• AI Assistant: Chat-based refinement interface
• Visual-First Mode: Image-centric slide layouts

👥 Team & Resources

Core Developers:

• Hrishikesh Jadhav (University of Passau)
• Javad Razavian (University of Qom)
• Moiz Khan Sherwani (University of Copenhagen)

Project Links:

• Code Repository
• Demo Video
• Hosted Application: AutoDeckAI Cloud

📜 License

MIT Licensed - View Full Terms

---

Empowering 1000+ Ecologists Worldwide
Report Issues • Request Features • Cite Our Work