A multimodal AI system that autonomously converts static PDF documents into professional PowerPoint presentations with synchronized, AI-generated audio narration.

Powered by an 8-stage agentic pipeline orchestrated with LangGraph and presented through a sleek Streamlit web interface, this tool delivers complete end-to-end automation.

The Automated Multimodal Agent is an advanced artificial intelligence system designed to autonomously convert static PDF documents into dynamic, narrated PowerPoint presentations. Utilizing a sophisticated multi-agent orchestration framework, the system breaks down the complex task of presentation creation into specialized micro-tasks. By seamlessly integrating document parsing, natural language processing (NLP), vector-based semantic retrieval, and Text-to-Speech (TTS) synthesis, the project significantly reduces the time and manual effort required to create professional, accessible, and engaging multimedia presentations.

You can explore it here: 🔗 https://medium.com/@mahmoudalyosify/from-static-pdf-to-talking-slides-automated-multimodal-agent-9b62c9a992c3

You can explore the full project details here:

🔗 https://mahmoudalyosifysite.github.io/Projects/Automated%20MultimodalAgent%20PDF%20to%20Narrated%20PowerPoint.html

▶️ Click the image above to watch the full demo on YouTube

• 📄 Intelligent PDF Parsing
  Extracts text, layout, and structural information seamlessly.

• 🧩 Semantic Chunking
  Organizes content into meaningful blocks for logical slide flow.

• 🔍 Vector-Based Retrieval
  Utilizes FAISS embeddings for intelligent content mapping.

• 🎨 Automated Slide Planning
  Designs optimal slide layouts and content distribution.

• ✍️ Content Generation
  Synthesizes dense text into concise, professional presentation points.

• 🔊 AI Voice Narration
  Generates natural-sounding TTS audio scripts and synchronized .wav files.

• 🌐 Professional Web UI
  Real-time progress tracking, error handling, and file management via Streamlit.

The system deviates from monolithic scripts by utilizing an Agentic Workflow.
Eight specialized agents handle specific micro-tasks, all coordinated by a master LangGraph orchestrator.

┌─────────────────────────────────────────────────────────┐
│                    Streamlit Web UI                     │
│  (Upload PDF → Monitor Progress → Download PPTX+Audio)  │
└────────────────────┬────────────────────────────────────┘
                     ↓
┌─────────────────────────────────────────────────────────┐
│              LangGraph StateGraph Orchestrator          │
├─────────────────────────────────────────────────────────┤
│  [1] PDF Parser   → Extract blocks                      │
│  [2] Chunker      → Create semantic chunks              │
│  [3] Vector DB    → Generate embeddings (FAISS)         │
│  [4] Planner      → Design slide layout                 │
│  [5] Generator    → Synthesize presentation text        │
│  [6] Script Agent → Write narration script              │
│  [7] TTS Agent    → Generate audio (.wav)               │
│  [8] PPTX Builder → Assemble final PowerPoint           │
└─────────────────────────────────────────────────────────┘

• Python 3.10+
• 8GB+ RAM (16GB recommended)
• 5GB Disk Space (for dependencies and local models)

Clone the repository and set up your virtual environment:

git clone https://github.com/yourusername/Automated-Multimodal-Agent-PDF-to-Narrated-PowerPoint.git
cd Automated-Multimodal-Agent-PDF-to-Narrated-PowerPoint

# Create and activate virtual environment
python -m venv venv

# On macOS / Linux
source venv/bin/activate  

# On Windows
venv\Scripts\activate

# Install dependencies
cd orchestrator
pip install -r requirements.txt

Launch the Streamlit application:

streamlit run app.py

Then open your browser:

http://localhost:8501

Upload your PDF, click Generate, and download your narrated PowerPoint when complete!

If you prefer the CLI version:

python "Agentic Systems/Master Orchestrator Agent/master_agent.py"

Frameworks

• LangGraph
• Streamlit
• Pydantic

NLP & ML

• HuggingFace Transformers
• PyTorch
• Sentence-Transformers
• scikit-learn

Document Processing

• PyMuPDF (fitz)
• python-pptx

Vector Search

• FAISS
• NumPy

Audio Synthesis

• pyttsx3
• SoundFile

All generated files are neatly organized in the output/ directory, timestamped for your convenience:

output/
└── 20260223_120000/
    ├── YourDocument_Narrated.pptx
    ├── audio/
    │   ├── slide_1.wav
    │   └── slide_2.wav
    └── metadata/
        ├── chunks.json
        ├── embeddings.json
        └── logs.txt

⚡ Utilizing an NVIDIA GPU with CUDA improves embedding generation speeds by approximately 30%.

• Async execution for parallel pipeline stages
• Integration with advanced cloud TTS providers (ElevenLabs, Azure)
• Multi-language support and translation capabilities
• Custom corporate .potx template support
• Vision-Language Models (VLMs) for chart/image extraction

• Mahmoud Alyosify
• Mirna Embaby

Distributed under the MIT License. See LICENSE for more information.

For extensive documentation, please refer to:

• README_COMPLETE.md
• SYSTEM_COMPLETE.md
• orchestrator/ markdown files

If you find this project useful:

• ⭐ Star the repository
• 🐛 Open issues for bugs
• 🚀 Submit pull requests

Automate. Narrate. Elevate.