{
  "id": 16538801,
  "title": "From 430 MB to 13 MB: Building an Offline AI Model for Indian GST Invoices",
  "url": "https://urgent.news/2026/10/26/from-430-mb-to-13-mb-building-an-offline-ai-model-for-indian-gst",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-26T00:34:23.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/pybankers/from-430-mb-to-13-mb-building-an-offline-ai-model-for-indian-gst-invoices-7nl"
  },
  "original_language": "en",
  "account": "Tiny GST, a compact AI model, has been developed to extract structured data from Indian GST invoices directly on-device. This model, named Tiny GST, requires only 13.17 MB after optimization, making it highly efficient. It boasts impressive accuracy, achieving 64.2% entity F1 for unseen vendor layouts, 84.6% F1 for unit-price extraction, and 80% F1 for HSN/SAC extraction. Notably, it achieves 100% accuracy in CGST amount F1. The model also provides validated TallyPrime-formatted XML and includes invoice deduplication using supplier GSTIN, invoice number, and financial year. This project aims to reduce cloud costs, protect sensitive data, and ensure invoice processing even in areas with unreliable internet connections. Currently, it is an early developer release, with live TallyPrime import testing still pending. The project can be found on GitHub at https://github.com/sanprat/GstEngine and the full engineering article is available at https://pybankers.com/blog/from-430mb-to-13mb-offline-ai-gst-invoice. The developer behind Tiny GST is Sanprat, who can be found on PyBankers at https://pybankers.com.",
  "summary": "We built Tiny GST, a compact AI model designed to extract structured data from Indian GST invoices directly on-device. The model: Runs fully offline Uses only 13.17 MB after INT8 ONNX quantization Achieves 64.2% entity F1 on unseen vendor layouts Reaches 84.6% F1 for unit-price extraction Reaches 80% F1 for HSN/SAC extraction Achieves 100% CGST amount F1 Generates validated TallyPrime-formatted…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}