AI Receipt Scanner

The AI Receipt Scanner is a browser-based capture app for receipt photos, PDFs and vehicle mileage images. It gives workers a simple upload screen while the backend extracts structured supplier, VAT, fuel, line-item, project-address and mileage data for review and export.

Challenge

Small trade and service teams lose time retyping receipts, chasing staff for mileage photos and reconciling supplier spend after the fact. The app needed to accept fast phone uploads, handle construction and fuel receipts, keep records scoped to the current user session and export usable data without a heavy accounting system rollout.

Solution

The live app provides separate receipt and mileage tabs. Receipt uploads support JPEG, PNG and PDF files with multi-file queue handling, MIME validation, 10MB limits and an optional project address override. Mileage capture validates the vehicle registration, accepts odometer photos and stores the extracted reading against the vehicle.

AI Extraction

The receipt flow uses Google Document AI for OCR, then OpenAI extraction to return structured Irish construction and fuel receipt fields: merchant, date, totals, VAT, VAT rate, payment method, category, line items, fuel type, litres, price per litre, confidence score, project address and optional GPS coordinates from image metadata. It is a concrete example of an agentic admin workflow with extraction, validation and human review.

Review and Export

Recent receipt cards show merchant, amount, category, payment method, project address and confidence score. Users can view details, inspect the uploaded image, delete records, filter by date and category, view daily totals and export the filtered receipt data to CSV, including project and GPS columns.

Mileage Workflow

Mileage capture has its own upload queue, vehicle registration input, odometer OCR, recent mileage log cards, image preview and CSV export. This lets a field team collect vehicle readings from phone photos without manually typing odometer values into a spreadsheet.

Data Storage

Records are written to private JSONL storage and scoped to the current browser session unless a real auth layer supplies a user email. Uploaded images are saved separately, while fetch and delete actions check ownership before returning or removing records.

Outcome

The app turns receipt and odometer photos into structured operational data with a low-friction upload flow. It is suitable as a standalone tool or as a module inside a wider work-order, fleet, HRM or finance platform.