app.py

import streamlit as st
from transformers import BertTokenizer, BertForSequenceClassification
import torch

# Load the model and tokenizer from Hugging Face
def load_model():
    model_name = "pritamdeb68/BERTAIDetector"
    model = BertForSequenceClassification.from_pretrained(model_name)
    tokenizer = BertTokenizer.from_pretrained(model_name)
    return model, tokenizer

# Load the model and tokenizer
model, tokenizer = load_model()

# Create a Streamlit app
st.title("AI Text Detector")

# Get user input
user_input = st.text_area("Enter your text:", height=300, placeholder="Enter your text here (10-100 words). Example: 'This is a sample text for analysis.'")

# Make predictions
if st.button("Detect"):
    # Process user input
    if len(user_input.split()) < 10:
        st.error("Please enter at least 10 words.")
    elif len(user_input.split()) > 100:
        user_input = ' '.join(user_input.split()[:100])
        st.warning("Text truncated to 100 words.")
    
    inputs = tokenizer(user_input, return_tensors="pt")
    with torch.no_grad():
        logits = model(**inputs).logits
    predicted_class = torch.argmax(logits).item()
    confidence = torch.nn.functional.softmax(logits, dim=1)
    max_confidence = confidence.max().item()

    # Update predictions based on confidence levels
    if max_confidence > 0.50 and predicted_class == 0:
        st.write("This text is likely **Human-written** content.")
    elif max_confidence < 0.90:
        st.write("This text is likely a mix of **AI-generated** and **Human-written** content.")
    elif predicted_class == 1:
        st.write("This text is likely **AI-generated**.")
    else:
        st.write("This text is likely **Human Written**.")