🧪 XAScribe

AI-powered research assistant for X-ray Absorption Spectroscopy (XAS) analysis and manuscript generation using Google Gemini 2.5 Flash and RAG technology.

Features

• XAS Data Analysis: Upload and analyze experimental data with machine learning models
• Literature Processing: Process research papers for context-aware manuscript generation with intelligent caching
• AI Manuscript Generation: Create academic content with proper citations using Gemini 2.5 Flash
• Smart Caching: Automatically reuses processed papers from previous sessions - no need to re-upload
• Optional Uploads: Add new papers to existing collections or work with cached data only
• Interactive Workflow: Streamlined interface with automatic detection of existing data

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Get Google Gemini API Key

1. Visit Google AI Studio
2. Create an API key for Gemini
3. Keep it ready for the app interface

3. Run the Application

streamlit run app.py

4. Access the App

Open your browser to the URL shown in terminal (typically http://localhost:8501)

Usage

1. Enter API Key in the sidebar
2. Check Existing Data: App automatically detects papers from previous sessions
3. Upload Data (optional):
   • XAS data files (JSON, CSV, XLSX, TXT)
   • Additional research papers (individual TXT files or ZIP archive)
4. Process Data: Initialize with existing papers or add new ones to your collection
5. Generate Manuscript: Enter your research question and generate academic content

Smart Caching System

• First Run: Upload research papers and XAS data - system creates cache files
• Subsequent Runs: Papers are automatically detected and loaded from cache
• Adding Papers: Upload additional papers to expand your existing collection
• Cache Files: rag_cache.pkl (metadata) and embeddings_cache.pkl (vector embeddings)

Example Data

The project includes:

• example_xas_data/nmc_exp_xas.json - Sample XAS experimental data
• example_papers/ - 24 research papers about XAS studies on battery cathodes
• paper_metadata.csv - Citation metadata for proper referencing

File Structure

├── app.py                    # Main Streamlit application
├── rag.py                    # RAG system with FAISS vector search
├── Backend/Modeling.py       # XAS data analysis pipeline
├── requirements.txt          # Python dependencies
├── paper_metadata.csv        # Citation metadata
├── rag_cache.pkl             # Cached paper metadata (auto-generated)
├── embeddings_cache.pkl      # Cached vector embeddings (auto-generated)
├── converted_papers/         # Processed paper text files (auto-generated)
├── example_xas_data/         # Sample experimental data
└── example_papers/           # Sample research papers

Configuration

Set your API key via:

• Streamlit sidebar interface (recommended)
• Environment variable: export GOOGLE_API_KEY="your-key"

Technologies

• Frontend: Streamlit
• AI: Google Gemini 2.5 Flash
• Vector Search: FAISS (Facebook AI Similarity Search)
• Embeddings: BGE-large-en (sentence-transformers)
• ML: Scikit-learn, SHAP
• Materials Science: Pymatgen
• Caching: Python pickle for persistent storage

Troubleshooting

• API Key Error: Verify your Gemini API key is valid and has access to Gemini 2.5 Flash
• File Format: Use supported formats (JSON, CSV, XLSX, TXT, ZIP)
• Memory Issues: Process smaller batches for large datasets
• Cache Issues: Delete rag_cache.pkl and embeddings_cache.pkl to force re-processing
• Missing Papers: Check converted_papers/ folder for processed text files
• Model Not Found: Ensure your API key has access to the latest Gemini models

Cache Management

• Cache Location: All cache files are stored in the project root directory
• Force Refresh: Delete cache files to re-process papers with updated parameters
• Backup: Cache files can be backed up to preserve processed literature databases
• Sharing: Share rag_cache.pkl and embeddings_cache.pkl to distribute processed datasets

---

XAScribe v2.0 - Advanced AI Research Platform for Materials Science with Smart Caching