Layout-Aware Summarization of Annual Reports for Multilingual India (v1.0 Stable)

An intelligent, robust system to extract, analyze, and summarize Indian annual reports with multilingual support and high-performance document processing.

🚀 Version 1.0 Highlights

• Optimized Performance: Single-pass document analysis with localized caching and instant tab switching.
• Lazy Loading: App startup is nearly instant; heavy AI models are loaded only when requested.
• Improved Stability: Robust error boundaries prevent crashes during model loading or processing failures.
• Memory Management: Automatic garbage collection and session purging when switching documents.

🛠️ Features

• Advanced PDF Processing: High-fidelity text extraction using multiple engines (pdfplumber, PyPDF2).
• Layout Analysis: Page structure visualization with bounding box identification.
• Multilingual Support:
  • Automatic Indian language detection.
  • Summarization for Hindi, Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, and Malayalam.
  • IndicBART Support: High-performance translation and summarization using local AI models.
• Interactive Visuals: Plotly charts for section distribution and keyword frequency.
• Export Options: Professional DOCX, CSV, and Text reports.

🛠️ Installation & Setup

1. Clone & Environment

   git clone <repository-url>
   cd annual_report_summarizer
   python -m venv venv
   source venv/bin/activate  # atau .\venv\Scripts\activate
   

2. Dependencies

   pip install -r requirements.txt
   

3. Resources

   • Ensure tesseract is installed for OCR features.
   • The app will automatically download ~1.5GB of AI models on the first run of "Direct" multilingual summarization.

🏃 Usage

streamlit run app.py

🔍 Troubleshooting

• indic_nlp_resources: If you see errors about missing Indic resources, ensure the indic_nlp_resources folder is present in the root directory.
• Memory Issues: For large PDFs (>100 pages), use a machine with at least 8GB of RAM if using direct multilingual models.
• Model Vulnerability: If you see torch.load warnings, this stable version has been patched to use safetensors for all AI model loading.

📁 Project Structure

annual_report_summarizer/
├── src/
│   ├── pdf_processor.py         # Advanced PDF extraction & logic
│   ├── layout_analyzer.py       # Visual structure analysis
│   ├── multilingual_summarizer.py # BART-based Indian language support
│   ├── language_detector.py    # Auto-detection for 10+ languages
│   ├── translator.py           # IndicBART & Google fallbacks
│   └── summarizer.py           # Extractive & Abstractive engines
├── app.py                      # Main Streamlit interface
└── requirements.txt            # Project dependencies

📄 License

MIT License - 2025