app.py

import streamlit as st
from pyvi.ViTokenizer import tokenize
from src.services.generate_embedding import generate_embedding
import pymongo
import time
from src.indexing import indexData, SHEET_ID, SHEET_NAME
from langchain_openai import ChatOpenAI
from langchain.prompts import ChatPromptTemplate
import os

os.environ["OPENAI_API_KEY"] = "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"

# Connect DB
client = pymongo.MongoClient(
    "mongodb+srv://rag:p9vojYc9fafYwxE9@rag.xswi7nq.mongodb.net/?retryWrites=true&w=majority&appName=RAG"
)
db = client.rag
collection = db.questionAndAnswers

with st.expander('Dataset'):
    col1 , col2 = st.columns(2)
    with col1: 
        st.markdown(
            """
            <div style="display:flex; gap: 16px; align-items: center">
                <a style="font-size: 14px"
                    href="https://docs.google.com/spreadsheets/d/1MKB6MHgL_lrPB1I69fj2VcVrgmSAMLVNZR1EwSyTSeA/edit#gid=0">Link
                    question & answers</a>
            </div>
            """,
            unsafe_allow_html=True,
        )

    with col2:
        if st.button('Re-train'):
            placeholder = st.empty()
            placeholder.empty()
            placeholder.write('Training ...')
            indexData(SHEET_ID, SHEET_NAME)
            placeholder.write('Completed')
            


def generateAnswer(context: str, question: str):
    prompt = ChatPromptTemplate.from_messages(
        [
            (
                "user","""Trả lời câu hỏi của người dùng dựa vào thông tin có trong thẻ <context> </context> được cho bên dưới. Nếu context không chứa những thông tin liên quan tới câu hỏi, thì đừng trả lời và chỉ trả lời là "Tôi không biết". <context> {context} </context> Câu hỏi: {question}""",
            ),
        ]
    )
    messages = prompt.invoke({"context": context, "question": question});
    print(messages)
    chat = ChatOpenAI(model="gpt-3.5-turbo-0125", temperature=0.8)
    response = chat.invoke(messages)
    return response.content


def stream_response(answer: str):
    for word in answer.split(" "):
        yield word + " "
        time.sleep(0.03)


# Initialize chat history
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat messages from history on app rerun
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"], unsafe_allow_html=True)

# React to user input
if prompt := st.chat_input(""):
    tokenized_prompt = tokenize(prompt)

    # Add user message to chat history
    st.session_state.messages.append({"role": "user", "content": tokenized_prompt})

    # Display user message in chat message container
    with st.chat_message("user"):
        st.markdown(tokenized_prompt)

    embedding = generate_embedding(tokenized_prompt)
    results = collection.aggregate(
        [
            {
                "$vectorSearch": {
                    "queryVector": embedding,
                    "path": "question_embedding",
                    "numCandidates": 10,
                    "limit": 10,
                    "index": "vector_index",
                }
            }
        ]
    )

    posibleQuestions = ""
    context = ""
    question = ""
    index = 0
    for document in results:
        posibleQuestions = posibleQuestions + f"<li>{document['question']}</li>"
        context =context + "\n\n" + document['question'] + ": " + document['answer']
        if index == 0:
            question = document["question"]
        index = index + 1
    posibleQuestions = f"""<ol> <p style="font-weight: 600">Câu hỏi liên quan: </p> {posibleQuestions}</ol>"""

    answer = generateAnswer(context, prompt);
    response = f"""<p>{answer}</p>
                    {posibleQuestions}
                    """

    # Display assistant response in chat message container
    with st.chat_message("assistant"):
        st.markdown(response, unsafe_allow_html=True)
        # st.markdown(f"""<p style="font-weight: 600">Question: {question}</p>""", unsafe_allow_html=True)
        # st.write_stream(stream_response(answer))
        # st.markdown(posibleQuestions, unsafe_allow_html=True)

    # Add assistant response to chat history
    st.session_state.messages.append({"role": "assistant", "content": response})