app.py

import logging
import gradio as gr
from transformers import pipeline
from sentence_transformers import SentenceTransformer, util
import PyPDF2

# Set up logging with a dedicated file handler
logger = logging.getLogger('SupportBot')
logger.setLevel(logging.INFO)
# Remove any existing handlers to avoid conflicts
if logger.handlers:
    logger.handlers.clear()
# Create a file handler with append mode
handler = logging.FileHandler('support_bot_log.txt', mode='a')
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(message)s')
handler.setFormatter(formatter)
logger.addHandler(handler)

# Load models
qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad")
embedder = SentenceTransformer('all-MiniLM-L6-v2')

# Helper function to extract text from PDF
def extract_text_from_pdf(file_path):
    text = ""
    with open(file_path, "rb") as file:
        pdf_reader = PyPDF2.PdfReader(file)
        for page in pdf_reader.pages:
            text += page.extract_text() + "\n"
    return text

# Find the most relevant section in the document
def find_relevant_section(query, sections, section_embeddings):
    stopwords = {"and", "the", "is", "for", "to", "a", "an", "of", "in", "on", "at", "with", "by", "it", "as", "so", "what"}
    
    logger.info(f"Searching for relevant section for query: {query}")
    query_embedding = embedder.encode(query, convert_to_tensor=True)
    similarities = util.cos_sim(query_embedding, section_embeddings)[0]
    best_idx = similarities.argmax().item()
    best_section = sections[best_idx]
    similarity_score = similarities[best_idx].item()
    
    SIMILARITY_THRESHOLD = 0.4
    if similarity_score >= SIMILARITY_THRESHOLD:
        logger.info(f"Found relevant section using embeddings (score: {similarity_score})")
        handler.flush()  # Ensure log is written immediately
        return best_section
    
    logger.info(f"Low similarity ({similarity_score}). Falling back to keyword search.")
    query_words = {word for word in query.lower().split() if word not in stopwords}
    for section in sections:
        section_words = {word for word in section.lower().split() if word not in stopwords}
        common_words = query_words.intersection(section_words)
        if len(common_words) >= 2:
            logger.info(f"Keyword match found with common words: {common_words}")
            handler.flush()
            return section
    
    logger.info("No good match found. Returning default response.")
    handler.flush()
    return "I don’t have enough information to answer that."

# Process the uploaded file
def process_file(file, state):
    logger.info("Received file upload request")
    if file is None:
        logger.info("No file uploaded")
        handler.flush()
        return [("Bot", "Please upload a file.")], state
    
    file_path = file.name
    if file_path.lower().endswith(".pdf"):
        logger.info(f"Processing PDF file: {file_path}")
        text = extract_text_from_pdf(file_path)
    elif file_path.lower().endswith(".txt"):
        logger.info(f"Processing TXT file: {file_path}")
        with open(file_path, 'r', encoding='utf-8') as f:
            text = f.read()
    else:
        logger.error(f"Unsupported file format: {file_path}")
        handler.flush()
        return [("Bot", "Unsupported file format. Please upload a PDF or TXT file.")], state
    
    sections = text.split('\n\n')
    section_embeddings = embedder.encode(sections, convert_to_tensor=True)
    state['document_text'] = text
    state['sections'] = sections
    state['section_embeddings'] = section_embeddings
    state['current_query'] = None
    state['feedback_count'] = 0
    state['mode'] = 'waiting_for_query'
    state['chat_history'] = [("Bot", "File processed. You can now ask questions.")]
    logger.info(f"File processed successfully: {file_path}")
    handler.flush()
    return state['chat_history'], state

# Handle user input (queries and feedback)
def handle_input(user_input, state):
    if state['mode'] == 'waiting_for_upload':
        logger.info("User input received before file upload")
        state['chat_history'].append(("Bot", "Please upload a file first."))
        handler.flush()
    elif state['mode'] == 'waiting_for_query':
        query = user_input
        logger.info(f"User query: {query}")
        state['current_query'] = query
        state['feedback_count'] = 0
        context = find_relevant_section(query, state['sections'], state['section_embeddings'])
        if context == "I don’t have enough information to answer that.":
            answer = context
        else:
            result = qa_model(question=query, context=context)
            answer = result["answer"]
        state['last_answer'] = answer
        state['mode'] = 'waiting_for_feedback'
        state['chat_history'].append(("User", query))
        state['chat_history'].append(("Bot", f"Answer: {answer}\nPlease provide feedback: good, too vague, not helpful."))
        logger.info(f"Generated answer: {answer}")
        handler.flush()
    elif state['mode'] == 'waiting_for_feedback':
        feedback = user_input.lower()
        logger.info(f"User feedback: {feedback}")
        state['chat_history'].append(("User", feedback))
        if feedback == "good" or state['feedback_count'] >= 2:
            state['mode'] = 'waiting_for_query'
            if feedback == "good":
                state['chat_history'].append(("Bot", "Thank you for your feedback. You can ask another question."))
                logger.info("Feedback 'good' received. Ready for next query.")
            else:
                state['chat_history'].append(("Bot", "Maximum feedback iterations reached. You can ask another question."))
                logger.info("Max feedback iterations (2) reached. Ready for next query.")
            handler.flush()
        else:
            query = state['current_query']
            context = find_relevant_section(query, state['sections'], state['section_embeddings'])
            if feedback == "too vague":
                adjusted_answer = f"{state['last_answer']}\n\n(More details:\n{context[:500]}...)"
                logger.info("Feedback 'too vague'. Providing context.")
            elif feedback == "not helpful":
                adjusted_answer = qa_model(question=query + " Please provide more detailed information with examples.", context=context)['answer']
                logger.info("Feedback 'not helpful'. Re-searching with modified query.")
            else:
                state['chat_history'].append(("Bot", "Please provide valid feedback: good, too vague, not helpful."))
                logger.info(f"Invalid feedback received: {feedback}")
                handler.flush()
                return state['chat_history'], state
            state['last_answer'] = adjusted_answer
            state['feedback_count'] += 1
            state['chat_history'].append(("Bot", f"Updated answer: {adjusted_answer}\nPlease provide feedback: good, too vague, not helpful."))
            logger.info(f"Updated answer: {adjusted_answer}")
            handler.flush()
    return state['chat_history'], state

# Initial state
initial_state = {
    'document_text': None,
    'sections': None,
    'section_embeddings': None,
    'current_query': None,
    'feedback_count': 0,
    'mode': 'waiting_for_upload',
    'chat_history': [("Bot", "Please upload a PDF or TXT file to start.")],
    'last_answer': None
}

# Gradio interface
with gr.Blocks() as demo:
    state = gr.State(initial_state)
    file_upload = gr.File(label="Upload PDF or TXT file")
    chat = gr.Chatbot()
    user_input = gr.Textbox(label="Your query or feedback")
    submit_btn = gr.Button("Submit")
    log_file = gr.File(label="Download Log File", value="support_bot_log.txt")

    file_upload.upload(process_file, inputs=[file_upload, state], outputs=[chat, state])
    submit_btn.click(handle_input, inputs=[user_input, state], outputs=[chat, state]).then(lambda: "", None, user_input)

demo.launch(share=True)