app.py

import os
import psycopg2
import streamlit as st
from datetime import datetime
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_huggingface import HuggingFaceEndpoint
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain, RetrievalQA
from huggingface_hub import login

# Login to Hugging Face
login(token=st.secrets["HF_TOKEN"])

# Load FAISS index and ensure it only happens once
if 'db' not in st.session_state:
    st.session_state.db = FAISS.load_local(
        "faiss_index", 
        HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2'),
        allow_dangerous_deserialization=True
    )

# Use session state for retriever
retriever = st.session_state.db.as_retriever(
    search_type="mmr",
    search_kwargs={'k': 1}
)

# Define prompt template
prompt_template = """
### [INST]
Instruction: You are a Q&A assistant. Your goal is to answer questions as accurately as possible based on the instructions and context provided without using prior knowledge. You answer in FRENCH
Analyse carefully the context and provide a direct answer based on the context. If the user said Bonjour or Hello your only answer will be Hi! comment puis-je vous aider?
Answer in french only

{context}
Vous devez répondre aux questions en français.
### QUESTION:
{question}
[/INST]
Answer in french only
Vous devez répondre aux questions en français.
"""

repo_id = "mistralai/Mistral-7B-Instruct-v0.3"

# Load the model only once
if 'mistral_llm' not in st.session_state:
    st.session_state.mistral_llm = HuggingFaceEndpoint(
        repo_id=repo_id, 
        max_length=2048, 
        temperature=0.05, 
        huggingfacehub_api_token=st.secrets["HF_TOKEN"]
    )

# Create prompt and LLM chain
prompt = PromptTemplate(
    input_variables=["question"],
    template=prompt_template,
)
llm_chain = LLMChain(llm=st.session_state.mistral_llm, prompt=prompt)

# Create QA chain
qa = RetrievalQA.from_chain_type(
    llm=st.session_state.mistral_llm,
    chain_type="stuff",
    retriever=retriever,
    chain_type_kwargs={"prompt": prompt},
)

# PostgreSQL connection setup using secrets from Hugging Face Spaces
def create_connection():
    return psycopg2.connect(
        host=os.getenv("DB_HOST"),
        database=os.getenv("DB_NAME"),
        user=os.getenv("DB_USER"),
        password=os.getenv("DB_PASSWORD"),
        port=os.getenv("DB_PORT")
    )

def insert_feedback(rating, comment):
    try:
        conn = create_connection()
        with conn.cursor() as cur:
            timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
            cur.execute('INSERT INTO feedback (timestamp, rating, comment) VALUES (%s, %s, %s)', 
                        (timestamp, int(rating), comment))
            conn.commit()
        return True
    except Exception as e:
        st.error(f"Error inserting feedback: {e}")
        return False
    finally:
        if conn:
            conn.close()

# Streamlit interface with improved aesthetics
st.set_page_config(page_title="Alter-IA Chat", page_icon="🤖")

def chatbot_response(user_input):
    response = qa.run(user_input)
    return response

# Create columns for logos
col1, col2, col3 = st.columns([2, 3, 2])

with col1:
    st.image("Design 3_22.png", width=150, use_column_width=True)

with col3:
    st.image("Altereo logo 2023 original - eau et territoires durables.png", width=150, use_column_width=True)

# CSS for styling
st.markdown("""
    <style>
    .centered-text {
        text-align: center;
    }
    .centered-orange-text {
        text-align: center;
        color: darkorange;
    }
    .star-rating {
        display: flex;
        flex-direction: row-reverse;
        justify-content: center;
        cursor: pointer;
    }
    .star-rating input[type="radio"] {
        display: none;
    }
    .star-rating label {
        font-size: 2em;
        color: #ddd;
        padding: 0 5px;
        transition: color 0.3s;
    }
    .star-rating input[type="radio"]:checked ~ label {
        color: gold;
    }
    .star-rating input[type="radio"]:hover ~ label {
        color: gold;
    }
    </style>
    """, unsafe_allow_html=True)

st.markdown('<h3 class="centered-text">🤖 AlteriaChat 🤖 </h3>', unsafe_allow_html=True)
st.markdown('<p class="centered-orange-text">"Votre Réponse à Chaque Défi Méthodologique "</p>', unsafe_allow_html=True)

# Input and button for user interaction
user_input = st.text_input("You:", "")
submit_button = st.button("Ask 📨")

if submit_button:
    if user_input.strip() != "":
        bot_response = chatbot_response(user_input)
        st.markdown("### Bot:")
        st.text_area("", value=bot_response, height=300)

        # Add rating and comment section
        st.markdown("---")
        st.markdown("#### Rate the Response:")

        # Use Streamlit form for feedback
        with st.form(key='feedback_form'):
            rating = st.select_slider("Rating:", options=[1, 2, 3, 4, 5], value=3)
            comment = st.text_area("Your Comment:")

            submit_feedback_button = st.form_submit_button("Submit Feedback")

            if submit_feedback_button:
                if comment.strip() == "":
                    st.warning("⚠ Please provide a comment.")
                else:
                    # Store feedback in PostgreSQL
                    if insert_feedback(rating, comment):
                        st.success("Thank you for your feedback!")
                    else:
                        st.error("Failed to submit feedback.")
    else:
        st.warning("⚠ Please enter a message.")

# Motivational quote at the bottom
st.markdown("---")
st.markdown("La collaboration est la clé du succès. Chaque question trouve sa réponse, chaque défi devient une opportunité.")