INTELLECTUAL PROPERTY OF VIRUJ THAKUR BHAIYA WITH 37CGP (GICENCE)

Patent Studio with Intellectual Property, Entropy & BlalajiLast Paper

A beautiful, locally-hosted patent analysis and Q&A application built with Streamlit and Ollama. Upload patent links from Excel, download full documents, auto-summarize them, and chat with individual patents using retrieval-augmented generation (RAG).

Features

• 📊 Excel Batch Ingestion: Upload an Excel file with a list of patent URLs (requires url column)
• 📥 Auto-Download & Extract: Automatically fetch PDF documents from patent links and extract full text
• ✨ AI Summarization: Generate concise patent summaries using local Ollama LLM (supports any Ollama model)
• 💬 Patent Q&A Chat: Ask questions about individual patents; responses are grounded in the patent text using semantic retrieval
• 📁 Local Storage: All downloaded PDFs, extracted text, and indexes stored locally in data/patents/ folder
• ⬇️ Export Options: Download summaries as CSV or all PDFs as ZIP archive
• 🎨 Beautiful UI: Modern gradient background, top navigation, tabbed patent workspace
• 🔄 Patent Switching: Easy dropdown navigation between processed patents
• 🚫 No Sidebar: Clean interface with top navigation bar and prominent back button

Project Structure

Patent Summarizer/
├── app.py                      # Main Streamlit app (upload, process, view results)
├── functions.py               # Core utilities (download, extract, summarize, chat, retrieval)
├── pages/
│   └── patent_viewer.py      # Patent workspace (PDF, Summary, Chat tabs)
├── .streamlit/
│   └── config.toml           # Streamlit configuration
├── data/
│   └── patents/              # Local storage for processed patents
│       └── pat_[hash]/       # Per-patent folder
│           ├── document.pdf
│           ├── document.txt
│           ├── summary.txt
│           ├── chunks.json
│           └── index.json
├── requirements.txt          # Python dependencies
└── README.md                 # This file

Installation

Prerequisites

• Python 3.9+ (this project uses 3.13.12)
• Ollama installed and running locally on localhost:11434
  • Download from: https://ollama.ai
  • Default model: llama3:8b (see configuration below)

Setup

1. Clone the repository:

   git clone <repo-url>
   cd "Patent Summarizer"

2. Create and activate a Python virtual environment:

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure Ollama (optional):

   • The app uses llama3:8b by default
   • To use a different model, change the default in the UI or modify functions.py
   • Ensure your chosen model is pulled: ollama pull <model-name>

5. Start Ollama (in a separate terminal):

   ollama serve

6. Run the app:

   streamlit run app.py

   The app will open at http://localhost:8501

Usage

Step 1: Upload Patents

1. Go to the main Patent Studio page
2. Upload an Excel file with a url column containing patent links
3. (Optional) Change Ollama model name
4. (Optional) Check "Reprocess existing" to re-download/re-summarize existing patents
5. Click Process Patents

Example Excel structure:

| url                                              |
|--------------------------------------------------|
| https://patents.google.com/patent/US10000001B2  |
| https://patents.google.com/patent/US10000002B2  |

Step 2: View Processing Results

• ✅ Successful patents appear in the "Processed Patents" list with status (cached/done)
• ❌ Failed URLs are shown in the "Failed URLs" table with error messages
• Download summaries as CSV or all PDFs as ZIP

Step 3: Open Patent Workspace

1. Click Open next to any patent in the list
2. In the patent workspace, use the top navigation dropdown to switch between patents
3. Three tabs available:
   • PDF: View the full patent document
   • Summary: Read the AI-generated summary
   • Chat: Ask questions about the patent

Step 4: Chat with Patents

1. Go to the Chat tab
2. Type a question (e.g., "What problem does this patent solve?")
3. The system retrieves relevant patent excerpts and uses Ollama to generate an answer
4. You can expand "Retrieved context" to see which excerpts were used
5. Click Clear chat to reset conversation for the current patent

Architecture

Data Flow

Excel Upload
    ↓
URL Normalization & Deduplication
    ↓
Patent Link Extraction (Selenium)
    ↓
PDF Download & Local Storage
    ↓
Text Extraction (PyMuPDF)
    ↓
AI Summarization (Ollama)
    ↓
Persistent Local Artifacts

Chat/Retrieval Pipeline

User Question
    ↓
Text Chunking (with overlap)
    ↓
Sparse Vector Embedding (token-based)
    ↓
Semantic Retrieval (top-4 chunks)
    ↓
Prompt Composition (system + context + history)
    ↓
Ollama LLM Response

Key Functions

functions.py

• download_pdf(url, filename) - Download PDF from URL
• extract_text(pdf_path) - Extract text using PyMuPDF
• summarize_text(text, model) - Generate summary via Ollama
• chunk_text(text, chunk_size, overlap) - Split text into overlapping chunks
• ensure_retrieval_index(text, index_path, chunks_path) - Build or load semantic index
• retrieve_chunks(question, chunks, embeddings, top_k) - Find relevant chunks
• chat_with_patent(question, patent_text, ...) - Full RAG chat flow
• get_pdf_link(url) - Extract PDF link from patent page (Selenium)

app.py

• check_ollama() - Health check for Ollama connectivity
• parse_urls_from_excel(file_obj) - Validate and normalize Excel input
• process_patent_url(url, model_name, force_reprocess) - Single patent pipeline
• create_pdf_zip(results) - Export PDFs as ZIP

pages/patent_viewer.py

• Top navigation (back button, patent dropdown, model selector)
• PDF viewing (native Streamlit PDF viewer)
• Summary display
• Chat interface with per-patent history

Configuration

Ollama Model

Change the default model in the UI on the main page, or modify in functions.py:

def summarize_text(text, model="llama3:8b"):  # Change here
    ...

Available models: https://ollama.ai/library

Streamlit Settings

.streamlit/config.toml:

[client]
toolbarMode = "minimal"

This hides the Streamlit toolbar and Deploy button for a cleaner UI.

Temperature & Parameters

• Summarization: temperature=0.1 (factual, concise)
• Chat: temperature=0.0 (precise Q&A), top_p=0.85 (focused output)

Local Storage

All artifacts are stored in data/patents/{patent_id}/:

• document.pdf - Original PDF file
• document.txt - Extracted patent text
• summary.txt - AI-generated summary
• chunks.json - Text chunks for retrieval
• index.json - Sparse vector embeddings
• metadata.json - Processing metadata (if added in future)

Patent IDs are deterministic (SHA1 hash of URL), so re-running with the same patents reuses cached artifacts.

Troubleshooting

"Ollama is not reachable at localhost:11434"

• Ensure Ollama is running: ollama serve in a separate terminal
• Check it's accessible: curl http://localhost:11434/api/tags

"Excel must contain a 'url' column"

• Make sure your uploaded Excel file has a column named exactly url
• Whitespace and case-sensitivity matter

PDF shows blank or doesn't load

• Refresh the page (Ctrl+R or Cmd+R)
• Check that the PDF file exists locally in data/patents/{patent_id}/document.pdf

Chat responses are slow

• Ollama inference takes time (depends on model size and hardware)
• Look for the "Thinking over patent context..." spinner
• Consider using a smaller/faster model if available

Sidebar appears on reload

• This should not happen with the latest build
• If it does, clear browser cache or hard refresh (Ctrl+Shift+R)

Dependencies

Package	Purpose
streamlit	Web UI framework
streamlit[pdf]	PDF rendering component
pandas	Excel/CSV data handling
requests	HTTP downloads
pymupdf	PDF text extraction
openpyxl	Excel backend
selenium	Browser automation for link extraction

Known Limitations

1. Patent Link Extraction: Selenium-based scraping is fragile; some patent sites may block or change structure
2. Sparse Embeddings: Uses token-overlap similarity, not semantic embeddings (no ML model needed locally, but less accurate than dense embeddings)
3. Single-User: No authentication; designed for local/personal use
4. No Database: Relies on file storage; metadata not indexed
5. Token Limits: Very large patents may exceed Ollama's context window

Future Enhancements

• Support for dense embeddings (sentence-transformers)
• Patent database/SQLite storage for metadata queries
• Batch comparison across multiple patents
• Export chat conversations
• Multi-user authentication
• Cloud Ollama support
• REST API for programmatic access

License

[Specify your license here, e.g., MIT, Apache 2.0]

Contributing

Contributions welcome! Please:

1. Test locally before submitting
2. Follow PEP 8 style guidelines
3. Update README for new features
4. Add error handling and validation

Support

For issues or questions:

• Check the Troubleshooting section
• Review Ollama docs: https://ollama.ai/docs
• Check Streamlit docs: https://docs.streamlit.io

---

ON IDEAS & PROPERTY

PROPERTY CAN BE INTELLECTUAL, DIGITAL OR PHYSICAL PROPERTY. INTELLECTUAL PROPERTY IS CREATION OF MIND. AS PER OUR CURRENT (2026 CE)) LIMITED KNOWLEDGE OF UNIVERSE INTELLECTUAL PROPERTY SEPARATES HUMANS FROM NON HUMAN ANIMALS BECAUSE ANIMALS ONLY EAT, SLEEP, PLAY, FIGHT, REPRODUCE & DIE BUT HUMANS ARE MORE CONCIOUS THEN MOST ANIMALS THEY SEE VARIAOUS PROBLEMS THEY TRY TO SOLVE IT USING MIND. IDENATIFY PATTERNS OR ALSO SATISFY THEIR CURIOSITY.

BY THIS LOGIC INTELLECTUAL PROPERTY IS MORE IMPORTANT THEN PHYSICAL PROPERTY AND IT I THE BAISIS OF DIGITAL & PHYSICAL PREOPRTY. LIKE AC & MOTOR IS CREATETION OF MND & EFFORT OF MICHAEL FARADAY & TESLA.

Built with ❤️ using Streamlit + Ollama

INTELLECTUAL PROPERTY OF VIRUJ THAKUR BHAIYA WITH 37CGP (GICENCE)