app.py

import os
import streamlit as st
import faiss
import pickle
from groq import Groq
from datasets import load_dataset
from transformers import pipeline

# Initialize Groq API
client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

# Load datasets
healthcare_ds = load_dataset("harishnair04/mtsamples")
education_ds = load_dataset("ehovy/race", "all")
finance_ds = load_dataset("warwickai/financial_phrasebank_mirror")

# FAISS Index Setup
index = faiss.IndexFlatL2(768)
chat_history = []

# Streamlit UI Setup
st.set_page_config(page_title="AI Chatbot", layout="wide")
st.title("🤖 AI Chatbot (Healthcare, Education & Finance)")

# Sidebar for chat history
st.sidebar.title("📜 Chat History")
if st.sidebar.button("Download Chat History"):
    with open("chat_history.txt", "w") as file:
        file.write("\n".join(chat_history))
    st.sidebar.success("Chat history saved!")

# Chat Interface
user_input = st.text_input("💬 Ask me anything:", placeholder="Type your query here...")
if st.button("Send"):
    if user_input:
        # Determine dataset based on user query (Basic CAG Implementation)
        dataset = healthcare_ds if "health" in user_input.lower() else \
                  education_ds if "education" in user_input.lower() else \
                  finance_ds
        
        # RAG: Retrieve relevant data
        retrieved_data = dataset['train'][0]  # Simplified retrieval
        
        # Generate response using Llama via Groq API
        chat_completion = client.chat.completions.create(
            messages=[{"role": "user", "content": f"{user_input} {retrieved_data}"}],
            model="llama-3.3-70b-versatile"
        )
        response = chat_completion.choices[0].message.content
        
        # Save chat to FAISS and display
        chat_history.append(f"User: {user_input}\nBot: {response}")
        st.text_area("🤖 AI Response:", value=response, height=200)

# Display past chats
st.sidebar.write("\n".join(chat_history))

# Save chat history using pickle for persistence
def save_chat_history():
    with open("chat_history.pkl", "wb") as file:
        pickle.dump(chat_history, file)

def load_chat_history():
    global chat_history
    if os.path.exists("chat_history.pkl"):
        with open("chat_history.pkl", "rb") as file:
            chat_history = pickle.load(file)

load_chat_history()
if st.sidebar.button("Save Chat History"):
    save_chat_history()
    st.sidebar.success("Chat history saved permanently!")