app.py

from collections import deque
from langchain_community.vectorstores import Pinecone
from langchain_openai import OpenAIEmbeddings
from langchain_huggingface import HuggingFaceEndpoint
from langchain.prompts import PromptTemplate
from langchain.chains import RetrievalQA
from pinecone import Pinecone, ServerlessSpec
from langchain_pinecone import PineconeVectorStore
from dotenv import load_dotenv
from langchain import hub
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain.chains.retrieval import create_retrieval_chain
from langchain_openai import ChatOpenAI
from langchain.chains import LLMChain
from langchain_community.retrievers import BM25Retriever
from langchain_core.runnables import RunnableLambda
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
import pinecone
import time
import requests
import os
import gradio as gr
import re
import gc
import tensorflow as tf
import openai
import base64

openai_api = os.getenv("open_api_key")
pinecone_api = os.getenv("pinecone_api_key")

index = "wooribank"

pc = Pinecone(api_key=pinecone_api)

#index = pc.Index("wooribank")

embeddings = OpenAIEmbeddings (
    model = "text-embedding-ada-002",
    openai_api_key = openai_api
)

llm = ChatOpenAI(
    openai_api_key=openai_api,
    model_name='gpt-4o',
    temperature=0.0
)

vectorstore = PineconeVectorStore(
    index_name="wooribank",
    embedding=embeddings,
    pinecone_api_key=pinecone_api,
    namespace="finance"
)

def search_documents(query, k=3):
    similarity_threshold = 0.8
    results = vectorstore.similarity_search_with_score(query, k=k)

    filtered_results = [doc for doc, score in results if score >= similarity_threshold]

    if not filtered_results:
        return None

    return "\n\n".join([doc.page_content for doc in filtered_results])

prompt = PromptTemplate(
    input_variables=["context", "question"],
    template=(
        "You are a helpful Korean AI assistant that answers financial questions based on the provided information.\n\n"
        "### Context ###\n"
        "{context}\n\n"
        "### Question ###\n"
        "{question}\n\n"
        "### Answer ###\n"
        "Based on the provided context, provide a relevant and helpful answer in Korean."
        "Your response should be concise, with no more than 5 sentences."
        "If the context provides relevant information, answer the question clearly without adding unnecessary details."
        "If and only if the context provides no relevant information at all, then respond with: '대답하기에 충분한 정보를 알지 못합니다. 다른 질문을 해주시겠어요?'."
        "If the context contains Woori Bank product details, and the query is not directly about Woori Bank, mention the product naturally at the end as an example."
    )
)

qa_chain = (
    {
        "context": RunnableLambda(search_documents),
        "question": RunnablePassthrough(),
    }
    | prompt
    | llm
    | StrOutputParser()
)

def image_to_base64(file_path):
    with open(file_path, "rb") as file:
        return base64.b64encode(file.read()).decode("utf-8")

#image_base64 = image_to_base64("https://huggingface.co./spaces/fidelkim/financial-advisory-for-students/resolve/main/temp.png")
loc = os.path.join(os.path.dirname(__file__), "temp.png")
image_base64 = image_to_base64(loc)

#logo = os.path.join(os.path.dirname(__file__), "logoeng.png")
#logo_base64=image_to_base64(logo)

css = f"""
.gradio-container {{
    background-image: url('data:image/jpeg;base64,{image_base64}');
    background-size: 700px 300px;
    background-position: left bottom;
    background-repeat: no-repeat;
}}

#custom-button {{
    background-color: #0067ac; 
    color: white;
    border: none; 
    border-radius: 5px; 
    padding: 10px 20px; 
    cursor: pointer;
}}

#custom-button:hover {{
    background-color: #2482C5;
}}

#chatbot-box {{
    max-height: 700px; 
    overflow-y: auto; 
    border: 1px solid #ccc; 
    border-radius: 10px; 
    padding: 10px;
    background-color: #f9f9f9; 
}}

#chatbot-box .message.user {{
    background-color: #2482C5; 
    color: white; 
    border-radius: 10px; 
    padding: 10px;
    text-align: right;
    box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
    animation: fadeIn 0.5s ease-in-out;
}}

#chatbot-box .message.bot {{
    background-color: #a0a0a0;
    color: white; 
    border-radius: 10px; 
    padding: 10px;
    text-align: left;
    box-shadow: 0px 4px 6px rgba(0, 0, 0, 0.1);
    animation: fadeIn 0.5s ease-in-out;
}}

@keyframes fadeIn {{
    from {{
        opacity: 0;
    }}
    to {{
        opacity: 1;
    }}
}}

"""

chat_history = []

def chatbot_interface(user_input):

    chat_history.append((user_input, "답변 생성 중..."))
    yield chat_history

    answer = qa_chain.invoke(user_input)

    temp = ""
    for char in answer:
        temp += char
        chat_history[-1] = (user_input, temp)
        yield chat_history
        time.sleep(0.05)

right_aligned_text = """
<div align="right">
    <h3>This project was developed as part of the Woori Bank internship program.</h3>
</div>
"""


with gr.Blocks(css=css) as demo:
    gr.Markdown("# Financial Helper Chatbot for Students")
    #gr.Markdown(title_with_logo)

    with gr.Row():
      with gr.Column():
        gr.Markdown("")
        gr.Markdown("")
        gr.Markdown("")
        gr.Markdown("# Your Nearest Financial Helper")
        gr.Markdown("## by Woori Bank")
        #gr.Image(logo)

      with gr.Column():
        gr.Markdown("")
        gr.Markdown("")
        chatbot = gr.Chatbot(label="대화 내역", elem_id="chatbot-box")
        user_input = gr.Textbox(label="", placeholder="메시지를 입력하세요...", lines=1, show_label=False)
        submit_button = gr.Button("전송", elem_id="custom-button")
        gr.Markdown("")
        #gr.Markdown("#### This project was developed as part of the Woori Bank internship.")
        gr.Markdown(right_aligned_text)

    submit_button.click(
        chatbot_interface,
        inputs = user_input,
        outputs=chatbot,
        queue=True
    )

    user_input.submit(
        chatbot_interface,
        inputs = user_input,
        outputs=chatbot,
        queue=True
    )


if __name__ == "__main__":
    demo.launch(debug=True)