A lightweight, GPU-free Retrieval-Augmented Generation (RAG) system for educational textbooks. It combines hierarchical chunking, metadata-enhanced sparse retrieval, and multimodal visualization for efficient, interpretable information retrieval.
- Structure-Aware Chunking: Preserves textbook hierarchy during content chunking.
- Metadata-Enriched Sparse Retrieval: Uses enhanced TF-IDF indexing for accurate search.
- Educational Query Reranking: Optimized for terminology, definitions, and contextual queries.
- Multimodal Results: Displays both text and relevant image-based retrieval results.
- Interactive UI: Clean, responsive interface built with Streamlit.
- Lightweight & Efficient: No vector database required, no external API dependencies, and fully explainable retrieval process.
- PDF Content Extraction: Extracts text and visual elements from educational materials.
- Hierarchical Subtopic Chunking: Chunks text while respecting the document's logical structure.
- Sparse TF-IDF Indexing: Computationally efficient indexing.
- Metadata-Aware Retrieval & Reranking: Advanced matching logic.
- Multimodal Visualization: Presenting contextually rich results.
Our ablation studies and retrieval performance metrics demonstrate high accuracy across various educational query types:
| Query Type | Top-3 Accuracy |
|---|---|
| Terminology Queries | 96% |
| Definition Queries | 92% |
| Section-Aware Queries | 95% |
| Long Educational Queries | 88% |
| Conceptual Paraphrase Queries | 71% |
- Python
- Scikit-Learn
- Streamlit
- NumPy & Pandas
- PDF Processing Libraries
Make sure you have Python 3.8+ installed.
- Clone the repository:
git clone <your-repo-url> cd Multimodal-Structure-Aware-Educational-RAG
- Install dependencies:
pip install -r requirements.txt
To start the interactive Streamlit interface:
streamlit run streamlit_app.pyContributions, issues, and feature requests are welcome! Feel free to check the issues page.
Made with by Nandeesh IB & Team