app.py

import gradio as gr
from sentence_transformers import SentenceTransformer, util
import openai  
import os

os.environ["TOKENIZERS_PARALLELISM"] = "false"

# Initialize paths and model identifiers for easy configuration and maintenance
filename = "output_topic_details.txt"  # Path to the file storing destress-specific details
retrieval_model_name = 'output/sentence-transformer-finetuned/'

# openai.api_key = os.environ["OPENAI_API_KEY"]

system_message = "You are a comfort chatbot specialized in providing information on destressing activities."
# Initial system message to set the behavior of the assistant
messages = [{"role": "system", "content": system_message}]
messages.append({
	"role": "system",
	"content": "Do not use Markdown Format. Do not include hashtags or asterisks"
})

# Attempt to load the necessary models and provide feedback on success or failure
try:
    retrieval_model = SentenceTransformer(retrieval_model_name)
    print("Models loaded successfully.")
except Exception as e:
    print(f"Failed to load models: {e}")

def load_and_preprocess_text(filename):
    """
    Load and preprocess text from a file, removing empty lines and stripping whitespace.
    """
    try:
        with open(filename, 'r', encoding='utf-8') as file:
            segments = [line.strip() for line in file if line.strip()]
        print("Text loaded and preprocessed successfully.")
        return segments
    except Exception as e:
        print(f"Failed to load or preprocess text: {e}")
        return []

segments = load_and_preprocess_text(filename)

def find_relevant_segment(user_query, segments):
    """
    Find the most relevant text segment for a user's query using cosine similarity among sentence embeddings.
    This version finds the best match based on the content of the query.
    """
    try:
        # Lowercase the query for better matching
        lower_query = user_query.lower()
        
        # Encode the query and the segments
        query_embedding = retrieval_model.encode(lower_query)
        segment_embeddings = retrieval_model.encode(segments)
        
        # Compute cosine similarities between the query and the segments
        similarities = util.pytorch_cos_sim(query_embedding, segment_embeddings)[0]
        
        # Find the index of the most similar segment
        best_idx = similarities.argmax()
        
        # Return the most relevant segment
        return segments[best_idx]
    except Exception as e:
        print(f"Error in finding relevant segment: {e}")
        return ""

def generate_response(user_query, relevant_segment):
    """
    Generate a response emphasizing the bot's capability in providing therapy, destressing activites, and student opportunities information.
    """
    try:
        user_message = f"Here's the information on your request: {relevant_segment}"

        # Append user's message to messages list
        messages.append({"role": "user", "content": user_message})
        
        response = openai.ChatCompletion.create(
            model="gpt-4",
            messages=messages,
            max_tokens=4000,
            temperature=0.5,
            top_p=1,
            frequency_penalty=0.5,
            presence_penalty=0.5,
        )
        
        # Extract the response text
        output_text = response['choices'][0]['message']['content'].strip()
        
        # Append assistant's message to messages list for context
        messages.append({"role": "assistant", "content": output_text})
        
        return output_text
        
    except Exception as e:
        print(f"Error in generating response: {e}")
        return f"Error in generating response: {e}"

def query_model(question):
    """
    Process a question, find relevant information, and generate a response.
    """
    if question == "":
        return "Welcome to CalmConnect's CalmBot! Ask me anything about destressing strategies and we'll provide you ways to unlock your inner calm!"
    relevant_segment = find_relevant_segment(question, segments)
    if not relevant_segment:
        return "Could not find specific information. Please refine your question or head to our resources page."
    response = generate_response(question, relevant_segment)
    return response

welcome_message = ""
topics = ""

theme = gr.themes.Default(
    primary_hue="neutral",
    secondary_hue="neutral",
).set(
    background_fill_primary='#6C7DA8',
    background_fill_primary_dark='#6C7DA8',
    background_fill_secondary="#f8f1ea",
    background_fill_secondary_dark="#f8f1ea",
    border_color_accent="#f8f1ea",
    border_color_accent_dark="#6C7DA8",
    border_color_accent_subdued="#f8f1ea",
    border_color_primary="#f8f1ea",
    block_border_color="#f8f1ea",
    button_primary_background_fill="#f8f1ea",
    button_primary_background_fill_dark="#f8f1ea"
)
    
# Setup the Gradio Blocks interface with custom layout components
with gr.Blocks(theme=theme) as demo:
    
    # Centering the content using Gradio Rows and Columns
    gr.Markdown(welcome_message)  # Display the formatted welcome message
          
        
    # Show the topics on the left side
    with gr.Row():
        with gr.Column():
            question = gr.Textbox(label="You", placeholder="What do you want to talk to CalmBot about?")
            answer = gr.Textbox(label="CalmBot's Response :D", placeholder="CalmBot will respond here..", interactive=False, lines=20)
            submit_button = gr.Button("Submit")
            submit_button.click(fn=query_model, inputs=question, outputs=answer)

demo.launch()

# Launch the Gradio app to allow user interaction
demo.launch(share=True)