app.py

# --LLMs
from langchain.chat_models import ChatOpenAI

# --chains, components
from langchain.chains import RetrievalQA
from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
)

# --embeddings
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.embeddings import CacheBackedEmbeddings

# --document Loaders, processing
from langchain.document_loaders.csv_loader import CSVLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter

# --vector Stores, storage
from langchain.vectorstores import FAISS
from langchain.storage import LocalFileStore

# --other ibraries
import chainlit as cl



system_template = """
Use the following pieces of context to answer the user's question.
You are an AI customer service assistant for Daysoff and respond to queries in Norwegian language (by default).
Your expertise is in providing fast, accurate and on-brand answers that covers frequently asked questions about 
Daysoff firmahytteordning and personvernspolicy.

If you don't know the answer, just say "Hmm, I'm not sure." Don't try to make up an answer.
If the question is not about either of these two topics, politely inform them that you are tuned to only answer 
questions about this and that all other queries are best directed to [email protected].

You can make inferences based on the context as long as it still faithfully represents the feedback.

Example of how your response should be, using a warm, direct, and professional tone:

```
foo
```

Begin!
----------------
{context}"""


messages = [
    SystemMessagePromptTemplate.from_template(system_template),
    HumanMessagePromptTemplate.from_template("{question}"),
]
prompt = ChatPromptTemplate(messages=messages)
chain_type_kwargs = {"prompt": prompt}



@cl.author_rename
def rename(orig_author: str):
    rename_dict = {"Just a moment": "Thinking.."}
    return rename_dict.get(orig_author, orig_author)

@cl.on_chat_start
async def start():

    msg = cl.Message(content="Building vector store...")
    await msg.send()

    loader = CSVLoader(file_path="./data/total_faq.csv", source_column="Answer")
    data = loader.load()

    for i, doc in enumerate(data):
        doc.metadata["row_index"] = i + 1
        doc.metadata["source"] = doc.metadata.get("Info_Url", "")

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    documents = text_splitter.transform_documents(data)

    store = LocalFileStore("./cache/")

    core_embeddings_model = OpenAIEmbeddings()
    embedder = CacheBackedEmbeddings.from_bytes_store(
        core_embeddings_model, store, namespace=core_embeddings_model.model
    )
    docsearch = await cl.make_async(FAISS.from_documents)(documents, embedder)

    chain = RetrievalQA.from_chain_type(
        ChatOpenAI(model="gpt-4", temperature=0.0, streaming=True),
        chain_type="stuff",
        return_source_documents=True,
        retriever=docsearch.as_retriever(),
        chain_type_kwargs={"prompt": prompt}
    )
  # "Kan jeg avbestille min reservasjon?"
    markdown_table = """
    ### Daysoff Firmahytteordning & Personvern FAQ
    
    | Type 🧐          | Populære spørsmål 👋🏼                                                         
    |------------------|-----------------------------------------------------------|
    | Firmahytteordning | • Jeg har lagt inn en bestilling. Hva skjer videre?      |               
    | Firmahytteordning | • Hva er betingelser for utleie?                         |                 
    | Firmahytteordning | • Kan jeg avbestille min reservasjon?                    |                  
    | Personvern        | • Med hvem deler dere mine personlige opplysninger?      |                
    | Personvern        | • Hvordan beskytter dere dataene mine?                   |                
    | Firmahytteordning | • Kan jeg ta med hund eller katt?                        |                 
    """


    msg = cl.Message(
        content=f"FAISS ready. Her er noen eksempler på spørsmål jeg kan svare på: 👇🏼\n\n{markdown_table}"
    )
    await msg.send()


    cl.user_session.set("chain", chain)

@cl.on_message
async def main(message):

    chain = cl.user_session.get("chain")

    cb = cl.AsyncLangchainCallbackHandler(
        stream_final_answer=True,
        answer_prefix_tokens=["FINAL", "ANSWER"]
    )

    cb.answer_reached = True

    res = await chain.acall(message, callbacks=[cb])
    return

    answer = res["result"]
    source_elements = []
    visited_sources = set()

    # --documents, user session
    docs = res.get("source_documents", [])
    metadatas = [doc.metadata for doc in docs]
    #all_sources = [m["source"] for m in metadatas]

    # --append source(s), specific rows only
    for doc, metadata in zip(docs, metadatas):
        row_index = metadata.get("row_index", -1)
        source = metadata.get("source", "")

    if row_index in [2, 8, 14] and source and source not in visited_sources:
        visited_sources.add(source)
        source_elements.append(
            cl.Text(content="https://www.daysoff.no" + source, name="Info_Url")
        )

    if source_elements:
        answer += f"\nSources: {', '.join([e.content for e in source_elements])}"
        await cl.Message(content=answer, elements=source_elements).send()
        #await cl.Message(content="Sources: " + ", ".join([e.content for e in source_elements])).send()
        return

    else:
        await cl.Message(content=f"No sources found").send()