app.py

# __import__('pysqlite3')
# import sys
# sys.modules['sqlite3'] = sys.modules.pop('pysqlite3')

# modify comment

import streamlit as st
from langchain.llms import Ollama
import os
import chromadb
from constants import CHROMA_SETTINGS
from langchain.chains import RetrievalQA
from langchain.embeddings import HuggingFaceEmbeddings
from chromadb.config import Settings

from langchain.vectorstores import Chroma
from langchain.llms import Ollama
from langchain.callbacks.base import BaseCallbackHandler

# Custom streamlit handler to display LLM outputs in stream mode
class StreamHandler(BaseCallbackHandler):
    def __init__(self, container, initial_text=""):
        self.container = container
        self.text=initial_text
    def on_llm_new_token(self, token: str, **kwargs) -> None:

        self.text+=token+"" 
        self.container.markdown(self.text)

# streamlit UI configuration
def setup_page():
    st.set_page_config(layout="wide")
    st.markdown("<h2 style='text-align: center; color: white;'>Your Personal MBA </h2>" , unsafe_allow_html=True)
    url = 'https://personalmba.com/'
    col1, col2, col3= st.columns(3)
    with col2:
        st.markdown("""
            <div style="text-align: center;">
            <h5 style='color: white;'>Inspired by </h5>
            <a href="%s">The Personal MBA by Josh Kaufman</a>  
            </div>
            """ % url, unsafe_allow_html=True)
    st.divider()

# get necessary environment variables for later use
def get_environment_variables():
    model = os.environ.get("MODEL", "mistral")
    embeddings_model_name = os.environ.get("EMBEDDINGS_MODEL_NAME", "all-MiniLM-L6-v2")
    persist_directory = os.environ.get("PERSIST_DIRECTORY", "db")
    target_source_chunks = int(os.environ.get('TARGET_SOURCE_CHUNKS', 4))
    return model, embeddings_model_name, persist_directory, target_source_chunks

# create knowledge base retriever
def create_knowledge_base(embeddings_model_name, persist_directory, target_source_chunks):
    embeddings = HuggingFaceEmbeddings(model_name=embeddings_model_name)
    db = Chroma(persist_directory=persist_directory, embedding_function=embeddings)
    retriever = db.as_retriever(search_kwargs={"k": target_source_chunks})
    return retriever

# handle query when user hit 'enter' on a question
def handle_query(query, model, retriever):
    with st.chat_message('assistant'):

        with st.spinner("Generating answer..."):
            message_placeholder = st.empty()
            stream_handler = StreamHandler(message_placeholder)  
            llm = Ollama(model=model, callbacks=[stream_handler])
            qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever, return_source_documents=False)
            res = qa(query)
            answer = res['result']
            message_placeholder.markdown(answer)
            return answer

# dictionary to store the previous messages, create a 'memory' for the LLM
def initialize_session():
    if 'messages' not in st.session_state:
        st.session_state.messages = []

# display the messages
def display_messages():
    for message in st.session_state.messages:
        with st.chat_message(message['role']):
            st.markdown(message['content'])

# example questions when user first load up the app. Will disappear after user send the first query
def show_examples():
    examples = st.empty()
    with examples.container():
        with st.chat_message('assistant'):
            st.markdown('Example questions:')
            st.markdown(' - How do I know that I am making the right decisions?')
            st.markdown(' - What are the key ideas in Chapter 6 "The Human Mind"?')
            st.markdown(' - What are common traits shared by the most sucessful individuals in the world?')
            st.markdown(' - I want to be a millionaire, build me a 5 year roadmap based on the top 0.01 percent of the human population.')
            st.markdown('So, how may I help you today?')
    return examples

    
def main():
    setup_page()
    initialize_session()
    display_messages()
    examples = show_examples()
    model, embeddings_model_name, persist_directory, target_source_chunks = get_environment_variables()
    retriever = create_knowledge_base(embeddings_model_name, persist_directory, target_source_chunks)
    
    query = st.chat_input(placeholder='Ask a question...')  # starting with empty query

    if query:   # if user input a query and hit 'Enter'
        examples.empty()

        st.session_state.messages.append({  # add the query into session state/ dictionary
            'role': 'user', 
            'content': query
        })

        with st.chat_message('user'):
            st.markdown(query)

        answer = handle_query(query, model, retriever)

        st.session_state.messages.append({  # add the answer into session state/ dictionary
             'role': 'assistant',
             'content': answer  
        })

if __name__ == "__main__":
    main()