chat/bot.py

# import streamlit as st
from langchain_pinecone.vectorstores import PineconeVectorStore
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFaceEndpoint
from langchain.prompts import PromptTemplate
from pinecone import Pinecone #, ServerlessSpec
from langchain_community.chat_message_histories import ChatMessageHistory
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
from langchain.retrievers import MergerRetriever
from dotenv import load_dotenv
import os
# from utils import process
from langchain_community.vectorstores import Chroma as LangChainChroma
import chromadb
# from chromadb.config import Settings
# from chromadb.utils import embedding_functions

# Load environment variables from the .env file
load_dotenv()

# Fetch environment variables
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
PINECONE_INDEX = os.getenv("PINECONE_INDEX")
HUGGINGFACE_API_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN")
# EMBEDDINGS_MODEL = os.getenv("EMBEDDINGS_MODEL")
# CHAT_MODEL = os.getenv("CHAT_MODEL")
EMBEDDINGS_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
CHAT_MODEL = "mistralai/Mixtral-8x7B-Instruct-v0.1"

# Supplement with streamlit secrets if None
# if None in [PINECONE_API_KEY, PINECONE_INDEX, HUGGINGFACE_API_TOKEN]: #, EMBEDDINGS_MODEL, CHAT_MODEL]:
#     PINECONE_API_KEY = st.secrets["PINECONE_API_KEY"]
#     PINECONE_INDEX = st.secrets["PINECONE_INDEX"]
    # HUGGINGFACE_API_TOKEN = st.secrets["HUGGINGFACEHUB_API_TOKEN"]
    # EMBEDDINGS_MODEL = st.secrets["EMBEDDINGS_MODEL"]
    # CHAT_MODEL = st.secrets["CHAT_MODEL"]

def ChatBot():
    embeddings = HuggingFaceEmbeddings(model_name=EMBEDDINGS_MODEL)
    # Initialize Pinecone
    pc = Pinecone(api_key=PINECONE_API_KEY)
    index = pc.Index(PINECONE_INDEX)
    pinecone_docsearch = PineconeVectorStore.from_existing_index(index_name=PINECONE_INDEX, embedding=embeddings)
    pinecone_retriever = pinecone_docsearch.as_retriever(
        search_kwargs={'filter': {'source': 'user_id'}}
    )
    chroma_client = chromadb.PersistentClient(path=":memory:")
    chroma_collection = chroma_client.get_or_create_collection(
        name="user_docs",
        # embedding_function=embeddings
    )
    langchain_chroma = LangChainChroma(
        client=chroma_client,
        collection_name="user_docs",
        embedding_function=embeddings
    )        
    
    # chroma_retriever = chroma_collection.as_retriever()
    chroma_retriever = langchain_chroma.as_retriever()
    
    # Combine retrievers
    combined_retriever = MergerRetriever(retrievers=[pinecone_retriever, chroma_retriever])

    # Initialize LLM and chain
    llm = HuggingFaceEndpoint(
        repo_id=CHAT_MODEL,
        model_kwargs={"huggingface_api_token":HUGGINGFACE_API_TOKEN},
        temperature=0.5,  ## make st.slider, subsequently
        top_k=10,  ## make st.slider, subsequently
    )

    prompt_template = """
        You are a trained bot to guide people about Illinois Crimnal Law Statutes and the Safe-T Act. You will answer user's query with your knowledge and the context provided. 
        If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
        Do not say thank you and tell you are an AI Assistant and be open about everything.
        Use the following pieces of context to answer the users question.
        Context: {context}
        Question: {question}
        Only return the helpful answer below and nothing else.
        Helpful answer:
        """

    PROMPT = PromptTemplate(
        template=prompt_template, 
        input_variables=["context", "question"])
    
    memory = ConversationBufferMemory(
        memory_key="chat_history",
        output_key="answer",
        chat_memory=ChatMessageHistory(),
        return_messages=True,
    )
    
    retrieval_chain = ConversationalRetrievalChain.from_llm(
        llm=llm,
        chain_type="stuff",
        retriever=combined_retriever,
        return_source_documents=True,
        combine_docs_chain_kwargs={"prompt": PROMPT},
        memory= memory
    )
    return retrieval_chain, chroma_collection, langchain_chroma