Indian GST Invoice Extractor using Ollama LLM

A powerful, privacy-first Streamlit application for extracting structured data from Indian GST invoices using Ollama's local vision models. No API keys required, runs completely offline, and outputs GST-compliant JSON format.

---

Key Features

GST Compliance

• Indian GST-compliant JSON output format
• Proper CGST, SGST, IGST rate and amount extraction
• HSN codes, PAN numbers, GST numbers extraction
• Tax summary with automated calculations
• Multiple items per invoice support

Privacy & Security

• 100% Local Processing - No data leaves your machine
• No API keys required - Uses local Ollama models
• Offline capable - Works without internet connection
• Cost-free - No per-request charges

File Format Support

• Images: PNG, JPG, JPEG
• Documents: PDF (first page extraction)
• Multi-file processing: Upload multiple invoices at once

Advanced Features

• Real-time processing with progress indicators
• Model selection - Choose from multiple vision models
• Data validation - Ensures GST compliance
• Export options: JSON, CSV summary, clipboard copy
• Error handling - Graceful failure recovery

---

Technologies Used

• Ollama: Local LLM inference with vision models

• Streamlit: Web interface

• Vision Models: LLaVA, BakLLaVA for image understanding

• Python Libraries:

  • ollama: Local LLM client
  • streamlit: Web interface
  • pandas: Data manipulation
  • Pillow: Image processing
  • PyMuPDF: PDF handling
  • base64: Image encoding

---

Installation

1. Clone the repository:

   git clone https://github.com/your-repository/invoice-extractor.git
   cd invoice-extractor

2. Install dependencies:

   pip install -r requirements.txt

3. Start the Ollama server:

   ollama serve

   (Ensure no other process is using port 11434. Close existing Ollama processes if needed.)

---

Usage

1. Run the Streamlit application:

   streamlit run app.py

2. Upload a file (PNG, JPG, PDF).

3. View extracted data in tabular format.

4. Export the data to your preferred format (CSV, JSON).

---

File Processing Workflow

1. Input Handling

   • Uploaded files are identified by type
   • Images are loaded using Pillow
   • PDFs are converted to images using fitz

2. Data Extraction

   • Ollama vision model processes the file content
   • Extracted data is parsed and formatted into a pandas DataFrame

3. Standardization

   • Dates are converted to YYYY-MM-DD format
   • Currency fields are formatted as INR with two decimals
   • Missing fields are replaced with None

4. Export Options

   • Download as CSV or JSON