Transaction Matching Microservice

Overview

This is a FastAPI-based microservice that performs machine learning-powered two-way matching between Open AR entries and Bank statements.

Features

• ML-based Matching: Uses BERT embeddings and TF-IDF vectorization for matching.

Installation

1. Clone the Repository

git clone <your-repo-url>
cd transaction-matching-service

2. Create and Activate Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

Running the Service

Start the FastAPI server by running:

uvicorn main:app --reload

API Endpoints

/match (POST)

• Description: Matches AR entries with Bank statements.
• Request Body:

{
    "ar_record": {
        "erpDocumentNumber": "INV123456",
        "documentDate": "2024-03-25",
        "customerNumber": "CUST001",
        "customerName": "ACME Corp",
        "paymentTerms": "Net 30",
        "dueDate": "2024-04-25",
        "invoiceAmount": 1500.0,
        "balanceAmount": 1500.0,
        "debitCreditIndicator": "D",
        "companyCode": "AC01"
    },
    "bank_statements": [
        {
            "id": 1,
            "transactionDate": "2024-03-26",
            "amount": 1500.0,
            "reference": "INV123456",
            "longText": "Payment from ACME Corp"
        },
        {
            "id": 2,
            "transactionDate": "2024-03-27",
            "amount": 1200.0,
            "reference": "INV789",
            "longText": "Partial payment"
        }
    ]
}

• Response:

🚀 Example Response

{
    "exact_match": [
        {
            "tfidf_similarity": 0.7092972666062738,
            "bert_similarity": 0.9373122453689575,
            "combined_similarity": 0.8233047559876157,
            "variance": 0.0,
            "variance_percentage": 0.0,
            "bank_statement_id": 1,
            "bank_reference": "INV123456",
            "bank_transaction_date": "2024-03-26",
            "bank_amount": 1500.0,
            "bank_long_text": "Payment from ACME Corp"
        }
    ],
    "possible_matches": [
        {
            "tfidf_similarity": 0.0,
            "bert_similarity": 0.6673458814620972,
            "combined_similarity": 0.3336729407310486,
            "variance": 300.0,
            "variance_percentage": 20.0,
            "bank_statement_id": 2,
            "bank_reference": "INV789",
            "bank_transaction_date": "2024-03-27",
            "bank_amount": 1200.0,
            "bank_long_text": "Partial payment"
        }
    ]
}

Future Improvements

• Add database integration to sync records between source systems
• Add endpoints for bank statement message types ingestion (MT940, BAI2, and other SWIFT messages)
• Add structured logging and monitoring

Dependencies

• FastAPI
• Sentence Transformers (BERT)
• Scikit-learn (TF-IDF Vectorizer)
• Uvicorn