app.py

import re
import gradio as gr
from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

# Load pretrained NER model
tokenizer = AutoTokenizer.from_pretrained("dslim/bert-base-NER")
model = AutoModelForTokenClassification.from_pretrained("dslim/bert-base-NER")
ner_pipeline = pipeline("ner", model=model, tokenizer=tokenizer, aggregation_strategy="simple")

# --- Core Redaction Function ---
def redact_text(input_text, custom_pattern=None, pattern_type=None):
    redacted_text = input_text

    # Apply model-based NER redaction
    ner_results = ner_pipeline(input_text)
    entities = sorted(ner_results, key=lambda x: x["start"], reverse=True)
    for ent in entities:
        redacted_text = (
            redacted_text[: ent["start"]] + "[REDACTED]" + redacted_text[ent["end"] :]
        )

    # Apply custom regex redaction if provided
    if pattern_type == "Date (YYYY-MM-DD or DD/MM/YYYY)":
        custom_pattern = r"\b\d{4}-\d{2}-\d{2}\b|\b\d{2}/\d{2}/\d{4}\b"
    elif pattern_type == "16-digit Number (e.g., Credit Card)":
        custom_pattern = r"\b\d{16}\b"

    if custom_pattern:
        redacted_text = re.sub(custom_pattern, "[REDACTED]", redacted_text)

    return redacted_text


# --- Gradio Interface ---
with gr.Blocks(title="PII Redactor") as demo:
    gr.Markdown("## 🔒 PII Redactor using BERT NER + Custom Regex")

    with gr.Row():
        text_input = gr.Textbox(
            label="Input Text",
            placeholder="Enter text containing PII (e.g. names, dates, locations, credit cards)...",
            lines=5,
        )

    with gr.Row():
        pattern_type = gr.Dropdown(
            ["None", "Date (YYYY-MM-DD or DD/MM/YYYY)", "16-digit Number (e.g., Credit Card)"],
            label="Select Regex Sample",
            value="None"
        )
        custom_pattern = gr.Textbox(
            label="Or Enter Custom Regex Pattern",
            placeholder=r"e.g. \b\d{3}-\d{2}-\d{4}\b for SSN",
        )

    redact_button = gr.Button("🔍 Redact PII")
    output_text = gr.Textbox(label="Redacted Output", lines=5)

    redact_button.click(
        fn=redact_text,
        inputs=[text_input, custom_pattern, pattern_type],
        outputs=output_text,
    )

# Launch the app
if __name__ == "__main__":
    demo.launch()