🏦 Bank Statements Analyzer

A Streamlit-based web app that uses LLama 3.2 Vision to extract structured financial data from bank statement images and return it in valid JSON — including customer info and transaction tables.

---

🚀 Features

• Upload a bank statement image (.jpg, .jpeg, .png)
• Uses LLama Vision via ollama.chat to extract data
• Returns strictly valid JSON in a predefined schema
• Parses transactions into a pandas DataFrame
• Displays extracted JSON and table in UI

---

📦 Installation

git clone <repo-url>
cd bank-statements-analyzer
pip install -r requirements.txt

▶️ Run the App

streamlit run bank_statements_analyzer.py

📤 Expected JSON Schema

The model is instructed to return only:

{
  "Name": "string",
  "Account Number/Customer Number": "string",
  "Account IBAN": "string",
  "Currency": "string",
  "Branch":"string or null",
  "Start Date": "YYYY-MM-DD",
  "End Date": "YYYY-MM-DD",
  "Transactions": [
    {
      "Date": "YYYY-MM-DD",
      "Transaction Details": "string",
      "Ref. No.":"string",
      "Deposit/Debit": "number or null",
      "Withdrawal/Credit": "number or null",
      "Balance": "number",
      "Value/Notes": "string or null"
    }
  ]
}

🛠 Tech Stack

Python

Streamlit

Ollama + Llama Vision

Pandas

📌 Notes

If extraction returns malformed JSON, the app shows raw model output.

Statements with low scan quality may reduce accuracy.

Made with ❤️ by Abdallah Ibrahim