app.py

import torch
import pandas as pd
from transformers import BartTokenizer, BartForConditionalGeneration
import gradio as gr

# Initialize models and tokenizers for Healthcare and AI perspectives
healthcare_model_name = 'facebook/bart-large-cnn'  # Healthcare-focused model
ai_model_name = 'facebook/bart-large-xsum'  # AI-focused model

healthcare_tokenizer = BartTokenizer.from_pretrained(healthcare_model_name)
ai_tokenizer = BartTokenizer.from_pretrained(ai_model_name)

healthcare_model = BartForConditionalGeneration.from_pretrained(healthcare_model_name)
ai_model = BartForConditionalGeneration.from_pretrained(ai_model_name)

# Summarization function for both Healthcare and AI agents
def generate_summary(text, tokenizer, model):
    inputs = tokenizer(text, return_tensors="pt", max_length=1024, truncation=True, padding="max_length")
    with torch.no_grad():
        outputs = model.generate(inputs["input_ids"], max_length=150, num_beams=5, no_repeat_ngram_size=2, early_stopping=True)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def healthcare_agent(abstract):
    return generate_summary(abstract, healthcare_tokenizer, healthcare_model)

def ai_agent(abstract):
    return generate_summary(abstract, ai_tokenizer, ai_model)

# Interaction function to generate implications based on both agents' insights
def generate_implications(healthcare_summary, ai_summary):
    healthcare_implication = f"Healthcare Implications: {healthcare_summary} The healthcare sector can leverage these findings to improve patient care and treatment outcomes."
    ai_implication = f"AI Implications: {ai_summary} These insights can further enhance AI models, making them more applicable in real-world healthcare scenarios."

    # Combine both implications to provide a holistic view
    combined_implications = f"{healthcare_implication}\n\n{ai_implication}"
    return combined_implications

# Function to process the CSV and generate results
def process_and_generate_implications(csv_file):
    # Read the input CSV file containing titles and abstracts
    papers_df = pd.read_csv(csv_file.name, encoding='latin-1')

    # Check if 'title' and 'abstract' columns exist
    required_columns = ['title', 'abstract']
    if not all(col.lower() in papers_df.columns.str.lower() for col in required_columns):
        return "The CSV must contain 'title' and 'abstract' columns."

    # Drop rows where title or abstract is missing
    papers_df = papers_df.dropna(subset=['title', 'abstract'])

    results = []

    # Process each paper (row) in the CSV
    for _, row in papers_df.iterrows():
        title = row['title']
        abstract = str(row['abstract'])

        # Generate summaries using both agents
        healthcare_summary = healthcare_agent(abstract)
        ai_summary = ai_agent(abstract)

        # Generate the implications based on both summaries
        implications = generate_implications(healthcare_summary, ai_summary)

        # Store the results
        results.append({
            "Title": title,
            "Abstract": abstract,
            "Healthcare Summary": healthcare_summary,
            "AI Summary": ai_summary,
            "Implications": implications
        })

    # Convert results into a DataFrame
    results_df = pd.DataFrame(results)

    # Return the results as a CSV string for download
    return results_df.to_csv(index=False)

# Define Gradio interface
def create_interface():
    with gr.Blocks() as demo:
        gr.Markdown("## Research Paper Summarization and Implications")
        gr.Markdown("Upload a CSV file with 'title' and 'abstract' columns to generate healthcare and AI implications.")

        # Upload CSV file
        csv_input = gr.File(label="Upload CSV File", type="file")

        # Button to process the CSV and generate results
        output_csv = gr.File(label="Download Results CSV")

        # Process CSV and generate implications on button click
        csv_input.change(process_and_generate_implications, inputs=csv_input, outputs=output_csv)

    return demo

# Launch the interface
if __name__ == "__main__":
    demo = create_interface()
    demo.launch(debug=True)  # Set debug=True to see detailed logs