app.py

import pandas as pd
import gradio as gr
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.schema import Document

# Initialize once when the app starts
def initialize_system():
    # Load dataset
    data = pd.read_csv("qa_dataset.csv")
    
    # Create documents
    documents = [
        Document(
            page_content=f"Q: {row['Question']}\nA: {row['Answer']}",
            metadata={"question": row['Question'], "answer": row['Answer']}
        ) for _, row in data.iterrows()
    ]
    
    # Create vector store
    embeddings = HuggingFaceEmbeddings(
        model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1"
    )
    
    return FAISS.from_documents(documents, embeddings)

vector_store = initialize_system()

def classify_question(query: str, k: int = 3):
    # Retrieve similar Q&A pairs
    results = vector_store.similarity_search(query, k=k)
    
    # Generate category from answers
    answers = " ".join([doc.metadata['answer'] for doc in results])
    keywords = list(dict.fromkeys(answers.split()))[:5]
    category = " ".join(keywords)
    
    # Format output
    return {
        "Category": category,
        "Top Matches": "\n\n".join([f"Q: {doc.metadata['question']}\nA: {doc.metadata['answer']}" 
                                  for doc in results]),
        "Confidence": f"{len(results)/k:.0%}"
    }

# Gradio interface
interface = gr.Interface(
    fn=lambda q: classify_question(q, 3),
    inputs=gr.Textbox(label="Input Question", placeholder="Type your question here..."),
    outputs=[
        gr.Textbox(label="Predicted Category"),
        gr.Textbox(label="Supporting Q&A"),
        gr.Textbox(label="Confidence")
    ],
    title="Question Classification System",
    description="Classify questions based on existing Q&A pairs using RAG"
)

if __name__ == "__main__":
    interface.launch()