ReceiptLens — AI-Powered Receipt Extraction & Expense Analytics

ReceiptLens is an AI-powered receipt information extraction and expense analysis application built using Google’s Gemma 4 E2B vision-language model, fine-tuned with LoRA (Low-Rank Adaptation) using Unsloth.

The application allows users to upload receipt images and automatically extract structured information, including the merchant name, date, address, total amount, receipt/transaction number, and contact number. Users can review and correct extracted fields, export the data as JSON or CSV, and analyze spending across multiple receipts through an interactive dashboard.

We fine-tuned Gemma 4 using the ICDAR 2019 SROIE dataset and evaluated it against the original model. On a 30-receipt test sample, complete extraction accuracy improved from 20% to 30%, with improvements in merchant, date, and total extraction.

Tech Stack: Python, Gemma 4 E2B, Unsloth, LoRA, Hugging Face, PyTorch, Google Colab, and Gradio.

Key Features:

• AI-powered receipt information extraction
• Fine-tuned multimodal vision-language model
• Editable extracted fields and JSON/CSV exports
• Receipt history and spending analytics dashboard
• Baseline vs. fine-tuned model evaluation
• Parameter-efficient training using LoRA

ReceiptLens demonstrates how open-weight AI models can be adapted for practical document understanding and financial data automation, while highlighting the importance of human verification and measurable model performance.