app.py

"""
import gradio as gr

def greet(name):
    return "Hello " + name + "!!"

demo = gr.Interface(fn=greet, inputs="text", outputs="text")
demo.launch()"""

import gradio as gr
import pandas as pd
import torch
from sentence_transformers import SentenceTransformer, util
import google.generativeai as genai
import pathlib
import textwrap
from IPython.display import display
from IPython.display import Markdown
import time

# 1. Initialization - Load Data and Set Up model
url = 'https://raw.githubusercontent.com/TRASJEPS/Vinicunca_AI_Project1/refs/heads/main/_Data-20241116T221029Z-001/Data/Cleaned%20Data/Combined%20Cleaned%20Data.csv'
df = pd.read_csv(url)

# Combine relevant columns
df["combined"] = (
    "Product Name: " + df.Product_Name.str.strip()+"; Brand: " + df.Brand.str.strip() 
    + "; Category: " + df.Product_Category.str.strip()
    + "; Details: " + df.Product_Details.str.strip()
    + "; Ingredients: " + df.Ingredients.str.strip()
    + "; Price: " + df["Cleaned Price"].str.strip()
    # +"; desc: "+ df.text.str.strip()
)

# Ensure the 'combined' column has no NaN or invalid values
df['combined'] = df['combined'].fillna('')

# Check that all values are strings
df['combined'] = df['combined'].astype(str)

# Convert the combined column to lowercase for consistency
df_combined = df.copy()
df_combined['combined'] = df_combined['combined'].str.lower()

# Load embedding model and move to GPU if available
model = SentenceTransformer('all-MiniLM-L6-v2')
if torch.cuda.is_available():
    modle = model.to('cuda')

# Create embeddings for all products once at the start
df['embeddings'] = df['combined'].apply(lambda x: model.encode(x))
df["embedding"] = df.combined.apply(lambda x: model.encode(x))

# 2. Define the Search Function for Gradio
def gradio_search(query):
    # Set the number of results to display
    n = 3

    # Embed the user query
    query_embedding = model.encode(query)

    # Calculate similarity
    df["similarity"] = df.embedding.apply(lambda x: util.cos_sim(x, query_embedding).item())

    # Sort by similarity and return the top 'n' results
    results = df.sort_values("similarity", ascending=False).head(n)
    resultlist = []

    # Collect results in a simple format
    for r in results.index:
        resultlist.append({
            "Product Name": results.Product_Name[r],
            "Score": results.similarity[r],
            "Category": results.Product_Category[r],
            "Price": results["Cleaned Price"][r],
            "Details": results.Product_Details[r],
            "Ingredients": results.Ingredients[r]
        })
    return resultlist

# 3. Maintain History for Multi-Turn Conversations
def chatbot_response(history, query):
    history.append(("User", query))  # Log user input
    results = gradio_search(query)
    
    if len(results) == 0:
        reply = "I couldn't find any matching products. Could you provide more details or rephrase your request?"
    else:
        reply = "\n".join([f"Product: {r['Product Name']}\nCategory: {r['Category']}\nPrice: {r['Price']}\nDetails: {r['Details']}\n" for r in results])

    history.append(("Vinuca", reply))  # Log bot response
    return history, reply

# 4. Set up the Gradio Interface
with gr.Blocks(title="Vinuca AI") as iface:
    gr.Markdown("# Vinuca AI, Your Personal Haircare Assistant")
    chatbot = gr.Chatbot()
    user_input = gr.Textbox(placeholder="Ask me about haircare products...")
    clear = gr.Button("Clear")

    history = []

    def interact(query):
        global history
        history, reply = chatbot_response(history, query)
        return history

    user_input.submit(interact, user_input, chatbot)
    clear.click(lambda: [], None, chatbot)

"""iface = gr.Interface(
    fn=gradio_search,
    inputs="text",
    outputs="json",
    title="Ulta Hair Recommendation Search",
    description="Enter your preferences to find matching products!"
)"""

# 4. Run the App
if __name__ == "__main__":
    iface.launch()