summerdevlin46/dukaan-saathi-receipt-lora

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 13, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

The summerdevlin46/dukaan-saathi-receipt-lora is a 3.2 billion parameter Llama-3.2-3B-Instruct fine-tune, specifically optimized for structured receipt parsing in Indian kirana (convenience) store workflows. This model excels at extracting detailed line items, quantities, prices, and supplier information from noisy, real-world receipt OCR text, including handwritten bills and informal notes. It is designed to convert unstructured receipt data into a structured JSON format, making it highly suitable for inventory management and data entry automation in this specific retail context.

Loading preview...

Overview

This model, summerdevlin46/dukaan-saathi-receipt-lora, is a fine-tuned version of Llama-3.2-3B-Instruct with 3.2 billion parameters, specifically developed for parsing receipts from Indian kirana (convenience) stores. It is part of the Dukaan Saathi inventory copilot demo.

Key Capabilities

  • Structured Data Extraction: Converts noisy OCR text from supplier receipts into a structured JSON object.
  • Handles Real-World Receipts: Designed to process various formats, including handwritten bills, printed tax invoices, and informal tally notes.
  • Detailed Line Item Parsing: Extracts product names, quantities (cases/units), unit costs, and total prices for each item.
  • Supplier and Invoice Information: Identifies supplier names, invoice numbers, dates, subtotals, discounts, and net totals.

Training and Limitations

  • Focused Training Data: Trained on a small dataset of 6 hand-authored and 22 Modal LLM-generated synthetic examples, totaling 28 examples. This focuses on format consistency for specific receipt styles.
  • Specific Use Case: Due to its specialized training, the model is not a general-purpose receipt parser and may overfit to known receipt styles (e.g., Mahalakshmi Marketing, Sri Venkateshwara Marketing, Brundavan Buns).
  • Human Oversight: Designed to work within a workflow that includes owner approval before any inventory write, acknowledging its specialized nature and potential for errors outside its trained scope.

When to Use This Model

  • Indian Kirana Store Automation: Ideal for automating inventory data entry from supplier receipts in the Indian convenience store context.
  • Structured Receipt Data: When you need to convert unstructured receipt text into a consistent, machine-readable JSON format for specific Indian retail scenarios.