Expense Receipt OCR Assistant

Problem

At month-end reimbursement, many WeChat Pay, Alipay, and bank transfer screenshots are scattered across phone galleries or chat threads. Manually copying amounts, dates, and merchants into a spreadsheet is error-prone and makes it hard to keep Excel headers consistent. This skill targets the workflow with many screenshots, fragmented fields, and inconsistent table formats by turning transaction screenshots into reusable reimbursement data.

Core Capabilities

• Screenshot recognition: supports WeChat Pay screenshots, Alipay bill screenshots, and bank transfer screenshots, and can also attempt invoice images and receipt photos.
• Field extraction: extracts key fields such as date, time, merchant/payee, amount, transaction type, and transaction number.
• Data cleansing: standardizes dates to YYYY-MM-DD, keeps amounts to two decimal places, and normalizes transaction types such as expense, income, and transfer.
• Spreadsheet generation: produces a standard Excel reimbursement table with columns including Date, Time, Merchant/Payee, Amount, Transaction Type, Transaction Number, and Notes.
• Batch processing: accepts multiple screenshots to reduce manual per-image cleanup.

How It Works

The skill relies on vision analysis for OCR and then invokes Excel processing to generate the file. The workflow has three main stages:

1. Use MiniMax VL API or GPT-4V to recognize screenshots and locate fields such as amount, date, merchant, and transaction type.
2. Clean extracted values: standardize dates, unify amount formatting, classify transaction types, and leave uncertain details in notes or for manual review.
3. Generate a normalized Excel file using the OpenXML SDK, producing a spreadsheet that can be reviewed, filtered, and submitted.

Boundaries and Caveats

It is suitable for personal or small-team reimbursement cleanup for daily expenses, travel, and transfers, but it is not a substitute for final finance review. If screenshots are blurry, key fields are obscured, one image contains multiple transactions, or the UI is non-standard, recognition accuracy may drop. Before use, configure a MiniMax API Key or OpenAI API Key, and ensure the vision analysis and Excel processing dependencies are available. After generation, still spot-check amounts, dates, and transaction numbers before submitting to finance systems.