AutoBank Intelligence

AI-powered banking transaction automation for intelligent financial data workflows. AutoBank Intelligence ingests bank transaction CSV files, trains a machine learning model to categorize transactions, detects unusual transactions, identifies recurring payments, and exposes the workflow through a FastAPI API.

This project uses synthetic sample data. Do not commit real banking data, account numbers, credentials, API keys, or customer information to GitHub.

Why this project is useful

This project demonstrates skills that are relevant for fintech, banking analytics, and data automation roles:

• CSV ingestion and validation
• Transaction categorization with machine learning
• Anomaly detection for unusual transactions
• Recurring payment detection
• Cashflow reporting
• FastAPI API design
• Testable Python project structure

Project structure

bankflow-automator/
├── app.py
├── data/
│   └── sample_transactions.csv
├── src/
│   └── bankflow/
│       ├── anomaly.py
│       ├── categorizer.py
│       ├── cli.py
│       ├── data_loader.py
│       ├── recurring.py
│       ├── reporting.py
│       └── tools.py
├── tests/
│   └── test_pipeline.py
├── requirements.txt
├── .gitignore
└── README.md

Setup

python -m venv .venv

Windows PowerShell:

.venv\Scripts\Activate.ps1

macOS/Linux:

source .venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run the command-line workflow

PYTHONPATH=src python -m bankflow.cli --file data/sample_transactions.csv

Save the output report:

PYTHONPATH=src python -m bankflow.cli --file data/sample_transactions.csv --output reports/report.json

On Windows PowerShell, use:

$env:PYTHONPATH="src"
python -m bankflow.cli --file data/sample_transactions.csv

Run the API

PYTHONPATH=src uvicorn app:app --reload

Then open:

http://127.0.0.1:8000/docs

Example request body for /analyze:

{
  "file_path": "data/sample_transactions.csv"
}

Main API endpoints

Endpoint	Method	Purpose
/health	GET	Check API status
/analyze	POST	Run the full banking automation workflow
/train-categorizer	POST	Train and evaluate the categorization model
/anomalies	POST	Detect unusual transactions

Data schema

The CSV file should include these columns:

Column	Meaning
transaction_id	Unique transaction identifier
date	Transaction date
description	Transaction text from bank statement
merchant	Merchant or counterparty
amount	Positive income or negative expense
currency	Currency code, for example EUR
balance	Account balance after transaction
account_id	Synthetic account identifier
label_category	Training label for the ML model

Example output

{
  "rows_analyzed": 153,
  "model_accuracy": 0.92,
  "detected_categories": ["Bank Fees", "Dining", "Entertainment", "Groceries"],
  "cashflow_report": {
    "total_income": 20180.0,
    "total_expenses": -13750.0,
    "net_cashflow": 6430.0
  },
  "top_anomalies": [],
  "recurring_payments": []
}

Suggested GitHub repository description

Machine learning banking automation project that categorizes transactions, detects anomalies, identifies recurring payments, and exposes the workflow through FastAPI.

Suggested next improvements

• Add Streamlit dashboard
• Add database storage with SQLite or PostgreSQL
• Add Plaid Sandbox or GoCardless Bank Account Data sandbox connector
• Add Dockerfile
• Add GitHub Actions CI
• Add time-series cashflow forecasting
• Add SHAP explanations for model interpretability