yakul259/credit-statement-scraper

💳 Credit Card Statement QA Model

Model Name: yakul259/credit-statement-scraper
Base Model: distilbert-base-uncased
Task: Question Answering (Extractive QA)
Framework: 🤗 Transformers
Language: English
Author: Yakul259
License: MIT

---

🧠 Model Overview

This model is a fine-tuned version of DistilBERT for question answering tasks, specifically designed to extract structured financial details from credit card statements in PDF or text format.

It was trained on a custom dataset of anonymized statements to recognize and answer questions like:

• “Which bank issued this statement?”
• “What is the billing cycle?”
• “What is the payment due date?”
• “What are the last 4 digits of the card?”
• “What is the total amount due?”

---

🏗️ Architecture

Property	Value
Model Type	DistilBERT
Architecture	DistilBertForQuestionAnswering
Hidden Size	768
Layers	6
Attention Heads	12
Max Sequence Length	512
Activation	GELU
Dropout	0.1
QA Dropout	0.1
Vocabulary Size	30,522
Transformers Version	4.57.0

---

🧾 Example Usage

You can load this model directly using the pipeline API from 🤗 Transformers:

from transformers import pipeline

qa_pipeline = pipeline(
    "question-answering",
    model="yakul259/credit-statement-scraper",
    tokenizer="yakul259/credit-statement-scraper"
)

context = """
Bank: XYZ Bank
Credit Card Number: **** **** **** 4321
Billing Period: 01/10/2025 - 31/10/2025
Payment Due Date: 15/11/2025
Total Amount Due: $1,254.67
"""

question = "What is the payment due date?"
result = qa_pipeline(question=question, context=context)
print(result)

## License
This model is released under the [MIT License](https://opensource.org/licenses/MIT).

### Attribution
This model was fine-tuned from [DistilBERT base uncased](https://huggingface.co/distilbert-base-uncased),
originally released by Hugging Face under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).

While the fine-tuned weights are distributed under the MIT License, users should note that the underlying
DistilBERT architecture and tokenizer originate from the Apache 2.0–licensed release.