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 chess-specific details
retrieval_model_name = 'output/sentence-transformer-finetuned/'

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

system_message = "You are a college chatbot specialized in providing information on college,scholarships, and mentors."
# Initial system message to set the behavior of the assistant
messages = [{"role": "system", "content": system_message}]

# 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):
    try:
        user_message = f"Here's what I found about scholarships: {relevant_segment}"
        messages.append({"role": "user", "content": user_message})
        
        response = openai.ChatCompletion.create(
            model="gpt-4o", 
            messages=messages,
            max_tokens=500, # can try increasing this if responses are cut off
            temperature=0.5,
            top_p=1,
            frequency_penalty=0.5,
            presence_penalty=0.5,
        )
        
        return response['choices'][0]['message']['content'].strip()
        
    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 ScholarSage! Ask me anything about college, scholarships, or mentors!"
    relevant_segment = find_relevant_segment(question, segments)
    if not relevant_segment:
        return "Sorry, I couldn't find this information! Please refine your question."
    response = generate_response(question, relevant_segment)
    return response

# Define the welcome message and specific topics the chatbot can provide information about
welcome_message = """
# ♟️ Welcome to ScholarSage!

## Your AI-driven assistant for all college-related queries. Created by Sadia, Jinny, and Kristy of the 2024 Kode With Klossy NYC Camp. 
"""

topics = """
### Feel Free to ask me anything from the topics below!
- College
- Scholarships
"""

subtopics = """
### Focus questions on these subtopics:
- List of Colleges in NYS
    - best colleges for CS
    - colleges that meet 100% financial need
    - private
    - public
    - community colleges
- List of Scholarships
    - low income student friendly
    - specific to a certain college
    - national scholarships


"""


# Setup the Gradio Blocks interface with custom layout components
with gr.Blocks(theme='JohnSmith9982/small_and_pretty') as demo:
    gr.Markdown(welcome_message)  # Display the formatted welcome message
    with gr.Column():
        gr.Markdown(topics)  # Show the topics on the left side
        gr.Markdown(subtopics)
    with gr.Column():   
        question = gr.Textbox(label="Your question", placeholder="What do you want to ask about?")
        answer = gr.Textbox(label="ScholarSage Response", placeholder="ScholarSage will respond here...", interactive=False, lines=10)
        submit_button = gr.Button("Submit")
        submit_button.click(fn=query_model, inputs=question, outputs=answer)
    

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