app.py

import chainlit as cl
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.document_loaders.csv_loader import CSVLoader
from langchain.embeddings import CacheBackedEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.chains import RetrievalQA
from langchain.chat_models import ChatOpenAI
from langchain.storage import LocalFileStore
from langchain.prompts.chat import (
    ChatPromptTemplate,
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
)
import chainlit as cl

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)

system_template = """
Use the following pieces of context to answer the user's question.
Please respond as if you are a customer support assistant ('kundeservice AI assistent') for Daysoff.
By default, you respond in Norwegian language (unless asked otherwise)
using a warm, direct, and professional tone.
Your expertise covers FAQs, and privacy policies.
If you don't know the answer, just say that you don't know, don't try to make up an answer and
politely redirect users to customer service at [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 direct:

```
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 = {"RetrievalQA": "Checking FAQ for ansatte & utleiere.."}
    return rename_dict.get(orig_author, orig_author)

@cl.on_chat_start
async def init():
    msg = cl.Message(content=f"Building Index...")
    await msg.send()

    # --builds FAISS index from csv
    loader = CSVLoader(file_path="./data/total_faq.csv", source_column="Answer") 
    data = loader.load()

    # --adding spec. metadata-------------------------------------------------------------------------------------------------
    for i, doc in enumerate(data):
        doc.metadata["row_index"] = i + 1  # --row index (1-based)
        doc.metadata["source"] = doc.metadata.get("Info_Url", "") 
    # ------------------------------------------------------------------------------------------------------------------------

    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
    )
    # --make async docsearch
    docsearch = await cl.make_async(FAISS.from_documents)(documents, embedder)

    chain = RetrievalQA.from_chain_type(
        ChatOpenAI(model="gpt-3.5-turbo", temperature=0.7, streaming=True),
        chain_type="stuff",
        return_source_documents=True,
        retriever=docsearch.as_retriever(),
        chain_type_kwargs = {"prompt": prompt}
    )

    msg.content = f"Index built! :)"
    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()
        return
        
    else:
       
        await cl.Message(content=f"No sources found").send()