app.py

from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.chains import  RetrievalQA
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import PyMuPDFLoader
import os
from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType
import chainlit as cl

@cl.on_chat_start
async def start():
    welcome_message1 = "Hi, I'm **Ben**! Your Benchmarking Assistant"
    welcome_message2 = "Powered by GPT-4, I can assist you with in-depth analysis of annual reports of some of the world's leading tech giants:\n- **Meta**\n- **Amazon**\n- **Alphabet**\n- **Apple**\n- **Microsoft**,\n\nfor the years **2022, 2021, and 2020**.\n\nWhether you are comparing financial performances, exploring trends, or seeking detailed insights across different years, Ben transforms complex data into actionable knowledge. Unlock the power of informed decision-making today with me!\n\nPlease ask a question to begin!"
    
    sample_questions="Some of the questions you can try:\n***"
    elements = [
        cl.Image(path="BenLogo2.png", name="Ben", display="inline", size="large"),
        # cl.Text(content=welcome_message1, name="10K-GPT", display="inline"),
    ]
    await cl.Message(content=welcome_message1, elements=elements).send()
    await cl.Message(content=welcome_message2).send()

@cl.langchain_factory(use_async=False)
def load():
    embeddings = OpenAIEmbeddings()
    llm = ChatOpenAI(temperature=0, model="gpt-4", streaming=True)

    apple_2022_docs_store = FAISS.load_local(r'data/datastores/apple_2022', embeddings)
    apple_2021_docs_store = FAISS.load_local(r'data/datastores/apple_2021', embeddings)
    apple_2020_docs_store = FAISS.load_local(r'data/datastores/apple_2020', embeddings)
    
    microsoft_2022_docs_store = FAISS.load_local(r'data/datastores/msft_2022', embeddings)
    microsoft_2021_docs_store = FAISS.load_local(r'data/datastores/msft_2021', embeddings)
    microsoft_2020_docs_store = FAISS.load_local(r'data/datastores/msft_2020', embeddings)

    amazon_2022_docs_store = FAISS.load_local(r'data/datastores/amzn_2022', embeddings)
    amazon_2021_docs_store = FAISS.load_local(r'data/datastores/amzn_2021', embeddings)
    amazon_2020_docs_store = FAISS.load_local(r'data/datastores/amzn_2020', embeddings)

    alphabet_2022_docs_store = FAISS.load_local(r'data/datastores/alphbt_2022', embeddings)
    alphabet_2021_docs_store = FAISS.load_local(r'data/datastores/alphbt_2021', embeddings)
    alphabet_2020_docs_store = FAISS.load_local(r'data/datastores/alphbt_2020', embeddings)

    meta_2022_docs_store = FAISS.load_local(r'data/datastores/meta_2022', embeddings)
    meta_2021_docs_store = FAISS.load_local(r'data/datastores/meta_2021', embeddings)
    meta_2020_docs_store = FAISS.load_local(r'data/datastores/meta_2020', embeddings)


    apple_2022_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=apple_2022_docs_store.as_retriever(search_kwargs={'k':5}))
    apple_2021_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=apple_2021_docs_store.as_retriever(search_kwargs={'k':5}))
    apple_2020_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=apple_2020_docs_store.as_retriever(search_kwargs={'k':5}))

    microsoft_2022_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=microsoft_2022_docs_store.as_retriever(search_kwargs={'k':5}))
    microsoft_2021_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=microsoft_2021_docs_store.as_retriever(search_kwargs={'k':5}))
    microsoft_2020_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=microsoft_2020_docs_store.as_retriever(search_kwargs={'k':5}))

    amazon_2022_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=amazon_2022_docs_store.as_retriever(search_kwargs={'k':5}))
    amazon_2021_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=amazon_2021_docs_store.as_retriever(search_kwargs={'k':5}))
    amazon_2020_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=amazon_2020_docs_store.as_retriever(search_kwargs={'k':5}))

    meta_2022_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=meta_2022_docs_store.as_retriever(search_kwargs={'k':5}))
    meta_2021_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=meta_2021_docs_store.as_retriever(search_kwargs={'k':5}))
    meta_2020_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=meta_2020_docs_store.as_retriever(search_kwargs={'k':5}))

    alphabet_2022_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=alphabet_2022_docs_store.as_retriever(search_kwargs={'k':5}))
    alphabet_2021_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=alphabet_2021_docs_store.as_retriever(search_kwargs={'k':5}))
    alphabet_2020_qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=alphabet_2020_docs_store.as_retriever(search_kwargs={'k':5}))



    tools = [
        Tool(
            name="Apple Form 10K 2022",
            func=apple_2022_qa.run,
            description="useful when you need to answer from Apple 2022",
        ),
        Tool(
            name="Apple Form 10K 2021",
            func=apple_2021_qa.run,
            description="useful when you need to answer from Apple 2021",
        ),
        Tool(
            name="Apple Form 10K 2020",
            func=apple_2020_qa.run,
            description="useful when you need to answer from Apple 2020",
        ),
        Tool(
            name="Microsoft Form 10K 2022",
            func=microsoft_2022_qa.run,
            description="useful when you need to answer from Microsoft 2022",
        ),
        Tool(
            name="Microsoft Form 10K 2021",
            func=microsoft_2021_qa.run,
            description="useful when you need to answer from Microsoft 2021",
        ),
        Tool(
            name="Microsoft Form 10K 2020",
            func=microsoft_2020_qa.run,
            description="useful when you need to answer from Microsoft 2020",
        ),
        Tool(
            name="Meta Form 10K 2022",
            func=meta_2022_qa.run,
            description="useful when you need to answer from Meta 2022",
        ),
        Tool(
            name="Meta Form 10K 2021",
            func=meta_2021_qa.run,
            description="useful when you need to answer from Meta 2021",
        ),
        Tool(
            name="Meta Form 10K 2020",
            func=meta_2020_qa.run,
            description="useful when you need to answer from Meta 2020",
        ),
        Tool(
            name="Alphabet Form 10K 2022",
            func=alphabet_2022_qa.run,
            description="useful when you need to answer from Alphabet or Google 2022",
        ),
        Tool(
            name="Alphabet Form 10K 2021",
            func=alphabet_2021_qa.run,
            description="useful when you need to answer from Alphabet or Google 2021",
        ),
        Tool(
            name="Alphabet Form 10K 2020",
            func=alphabet_2020_qa.run,
            description="useful when you need to answer from Alphabet or Google 2020",
        ),
        Tool(
            name="Amazon Form 10K 2022",
            func=amazon_2022_qa.run,
            description="useful when you need to answer from Amazon 2022",
        ),
        Tool(
            name="Amazon Form 10K 2021",
            func=amazon_2021_qa.run,
            description="useful when you need to answer from Amazon 2021",
        ),
        Tool(
            name="Amazon Form 10K 2020",
            func=amazon_2020_qa.run,
            description="useful when you need to answer from Amazon 2020",
        ),
    ]

    # Construct the agent. We will use the default agent type here.
    # See documentation for a full list of options.
    return initialize_agent(
        tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True
    )