app.py

# -*- coding: utf-8 -*-
"""ai_app.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1wUztAR4EdQUL3vkpM3Is-ps0TEocClry
"""

import gradio as gr
import pandas as pd
from langchain.document_loaders.csv_loader import CSVLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Qdrant
from langchain.chains import VectorDBQA
from langchain.llms import OpenAI


import os

openai_api_key = os.getenv('openai_api_key')
qdrant_url = os.getenv('QDRANT_URL')
qdrant_api_key = os.getenv('qdrant_api_key')



#csv loader
loader = CSVLoader(file_path='data.csv')
data=loader.load()

#split the documnts
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
texts = text_splitter.split_documents(data)


#embeding
embeding=OpenAIEmbeddings(openai_api_key=openai_api_key, model="text-embedding-3-small")

#import quantization

from langchain.vectorstores import Qdrant
from qdrant_client import QdrantClient, models

from langchain.vectorstores import Qdrant

#using qudadrant vector database
from qdrant_client import QdrantClient, models
qdrant = Qdrant.from_documents(
            texts,
            embeding,
            url=qdrant_url,
            prefer_grpc=True,
            api_key=qdrant_api_key,
            collection_name="llm_app_01",
            quantization_config=models.BinaryQuantization(
                binary=models.BinaryQuantizationConfig(
                    always_ram=True,
                )
            )
)

#qdrant client
qdrant_client = QdrantClient(
    url=qdrant_url,
    prefer_grpc=True,
    api_key=qdrant_api_key,
)

from re import search
#retriver
retriver=qdrant.as_retriever( search_type="similarity", search_kwargs={"k":2})


from langchain import PromptTemplate

prompt = PromptTemplate(
    template="""
        # Your Role
        You are a highly skilled AI specialized in healthcare and medical information retrieval. Your expertise lies in understanding the medical needs of patients and accurately matching them with the most suitable healthcare professionals based on the given context.

        # Instruction
        Your task is to answer the question using the following pieces of retrieved context delimited by XML tags.

        <retrieved context>
        Retrieved Context:
        {context}
        </retrieved context>

        # Constraint
        1. Carefully consider the user's question:
        User's question:\n{question}\n
        Analyze the intent behind the question, particularly in relation to the medical context, and provide a precise and helpful answer.
        - Reflect on why the question was asked and provide an appropriate response based on the context you understand.
        2. Select the most relevant information (the key details directly related to the question) from the retrieved context and use it to formulate an answer.
        3. Generate a concise, logical, and medically accurate answer. When generating the answer, include the following details about the doctor in a bulleted format:
            • Doctor Name: Dr. Shahzad Rashid Awan
            • City: Peshawar
            • Specialization: Dermatologist
            • Qualification: MBBS, MCPS (Dermatology)
            • Experience: 12 years
            • Patient Satisfaction Rate: 93%
            • Avg Time to Patients: 13 mins
            • Wait Time: 10 mins
            • Hospital Address: Rahim Medical Center And Hospital, Hasht Nagri, Peshawar
            • Fee: PKR 1000
            • Profile Link: https://www.marham.pk/doctors/peshawar/dermatologist/dr-shahzad-rashid-awan#reviews-scroll
        4. If the retrieved context does not contain information relevant to the question, or if the documents are irrelevant, respond with 'I can't find the answer to that question in the material I have'.
        5. Limit the answer to five sentences maximum. Ensure the answer is concise, logical, and medically appropriate.
        6. At the end of the response, provide the doctor's profile metadata as shown in the relevant documents, ensuring all bullet points are clearly mentioned.

        # Question:
        {question}""",
    input_variables=["context", "question"]
)


from langchain.chat_models import ChatOpenAI
llm = ChatOpenAI(model_name="gpt-4o", temperature=0, openai_api_key=openai_api_key)

def format_docs(docs):
        formatted_docs = []
        for doc in docs:
            # Format the metadata into a string
            metadata_str = ', '.join(f"{key}: {value}" for key, value in doc.metadata.items())

            # Combine page content with its metadata
            doc_str = f"{doc.page_content}\nMetadata: {metadata_str}"

            # Append to the list of formatted documents
            formatted_docs.append(doc_str)

        # Join all formatted documents with double newlines
        return "\n\n".join(formatted_docs)

from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
rag_chain = (
            {"context":  retriver| format_docs, "question": RunnablePassthrough()}
            | prompt
            | llm
            | StrOutputParser()
        )





from langchain.chat_models import ChatOpenAI
from langchain.schema import AIMessage, HumanMessage
import openai
import os
import gradio as gr

llm = ChatOpenAI(temperature=1.0, model='gpt-4o', openai_api_key=openai_api_key)

def reg(message, history):
    history_langchain_format = []
    for human, ai in history:
        history_langchain_format.append(HumanMessage(content=human))
        history_langchain_format.append(AIMessage(content=ai))
    history_langchain_format.append(HumanMessage(content=message))
    gpt_response = llm(history_langchain_format)
    return rag_chain.invoke(message)

gr.ChatInterface(reg).launch()