app.py

from langchain_community.document_loaders import PyPDFLoader,DirectoryLoader
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
import os
from langchain.prompts import PromptTemplate
from langchain_together import Together
from langchain.chat_models import ChatOpenAI
from htmlTemplates import css, bot_template, user_template
from langchain.embeddings import openai
from langchain.memory import ConversationBufferWindowMemory
from langchain.chains import ConversationalRetrievalChain
import streamlit as st
import time

from openai import OpenAI
api_key = os.getenv("OPENAI_API_KEY")
client = OpenAI(api_key=api_key)


# creating custom template to guide llm model
custom_template ="""<s>[INST]You will start the conversation by greeting the user and introducing yourself as qanoon-bot,\
stating your availability for legal assistance. Your next step will depend on the user's response.\
If the user expresses a need for legal assistance in Pakistan, you will ask them to describe their case or problem.\
After receiving the case or problem details from the user, you will provide the solutions and procedures according to the knowledge base and also give related penal codes and procedures. \
However, if the user does not require legal assistance in Pakistan, you will immediately thank them and\
say goodbye, ending the conversation. Remember to base your responses on the user's needs, providing accurate and\
concise information regarding the Pakistan legal law and rights where applicable. Your interactions should be professional and\
focused, ensuring the user's queries are addressed efficiently without deviating from the set flows.\
CHAT HISTORY: {chat_history}
QUESTION: {question}
ANSWER:
</s>[INST]
"""

embeddings=OpenAIEmbeddings()
#embeddings = HuggingFaceEmbeddings(model_name="nomic-ai/nomic-embed-text-v1",model_kwargs={"trust_remote_code":True,"revision":"289f532e14dbbbd5a04753fa58739e9ba766f3c7"})
#vectordb = Chroma.from_documents(texts, embedding=embeddings, persist_directory="./data")
#db_retriever =vectordb.as_retriever(search_type="similarity",search_kwargs={'k':4})

db = FAISS.load_local("vectordb", embeddings, allow_dangerous_deserialization=True)
db_retriever = db.as_retriever(search_type="similarity",search_kwargs={"k": 4})


st.set_page_config(page_title="Qanoon-Bot")
col1, col2, col3 = st.columns([1,4,1])
with col2:
    st.image("https://s3.ap-south-1.amazonaws.com/makerobosfastcdn/cms-assets/Legal_AI_Chatbot.png")

st.markdown(
    """
    <style>
div.stButton > button:first-child {
    background-color: #ffd0d0;
}
div.stButton > button:active {
    background-color: #ff6262;
}
   div[data-testid="stStatusWidget"] div button {
        display: none;
        }
    
    .reportview-container {
            margin-top: -2em;
        }
        #MainMenu {visibility: hidden;}
        .stDeployButton {display:none;}
        footer {visibility: hidden;}
        #stDecoration {display:none;}
    button[title="View fullscreen"]{
    visibility: hidden;}
        </style>
""",
    unsafe_allow_html=True,
)

def reset_conversation():
  st.session_state.messages = []
  st.session_state.memory.clear()

if "messages" not in st.session_state:
    st.session_state.messages = []

if "memory" not in st.session_state:
    st.session_state.memory = ConversationBufferWindowMemory(k=2, memory_key="chat_history",return_messages=True) 

#embeddings = HuggingFaceEmbeddings(model_name="nomic-ai/nomic-embed-text-v1",model_kwargs={"trust_remote_code":True,"revision":"289f532e14dbbbd5a04753fa58739e9ba766f3c7"})
#db=FAISS.load_local("/content/ipc_vector_db", embeddings, allow_dangerous_deserialization=True)


prompt = PromptTemplate(template=custom_template,
                        input_variables=['context', 'question', 'chat_history'])

# You can also use other LLMs options from https://python.langchain.com/docs/integrations/llms. Here I have used TogetherAI API

from config import together_api
llm=ChatOpenAI(temperature=0.2,model_name='gpt-3.5-turbo-0125')
qa = ConversationalRetrievalChain.from_llm(
    llm=llm,
    memory=st.session_state.memory,
    retriever=db_retriever,
    combine_docs_chain_kwargs={'prompt': prompt}
)

for message in st.session_state.messages:
    with st.chat_message(message.get("role")):
        st.write(message.get("content"))

input_prompt = st.chat_input("Say something")

if input_prompt:
    with st.chat_message("user"):
        st.write(input_prompt)

    st.session_state.messages.append({"role":"user","content":input_prompt})

    with st.chat_message("assistant"):
        with st.status("Thinking 💡...",expanded=True):
            result = qa.invoke(input=input_prompt)

            message_placeholder = st.empty()

            full_response = "**_Note: Information provided by Qanoon-Bot may be inaccurate. ** \n\n\n"
        for chunk in result["answer"]:
            full_response+=chunk
            time.sleep(0.02)
            
            message_placeholder.markdown(full_response+" ▌")
        st.button('Reset All Chat 🗑️', on_click=reset_conversation)

    st.session_state.messages.append({"role":"assistant","content":result["answer"]})