app.py

import os
import streamlit as st

from langchain_google_genai import GoogleGenerativeAIEmbeddings
from langchain_chroma import Chroma
from langchain_community.document_loaders import PyPDFLoader
from langchain_text_splitters.sentence_transformers import SentenceTransformersTokenTextSplitter
from langchain_core.prompts import ChatPromptTemplate
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough

from dotenv import load_dotenv
load_dotenv()

# Initialize embedding model
embedding_model = GoogleGenerativeAIEmbeddings(model="models/embedding-001")

# Initialize pharma database
db = Chroma(collection_name="pharma_database",
            embedding_function=embedding_model,
            persist_directory='./pharma_db')

def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

def add_to_db(uploaded_files):
    # Check if files are uploaded
    if not uploaded_files:
        st.error("No files uploaded!")
        return

    for uploaded_file in uploaded_files:
        # Save the uploaded file to a temporary path
        temp_file_path = os.path.join("./temp", uploaded_file.name)
        os.makedirs(os.path.dirname(temp_file_path), exist_ok=True)

        with open(temp_file_path, "wb") as temp_file:
            temp_file.write(uploaded_file.getbuffer())

        # Load the file using PyPDFLoader
        loader = PyPDFLoader(temp_file_path)
        data = loader.load()

        # Store metadata and content
        doc_metadata = [data[i].metadata for i in range(len(data))]
        doc_content = [data[i].page_content for i in range(len(data))]

        # Split documents into smaller chunks
        st_text_splitter = SentenceTransformersTokenTextSplitter(
            model_name="sentence-transformers/all-mpnet-base-v2",
            chunk_size=100,
            chunk_overlap=50
        )
        st_chunks = st_text_splitter.create_documents(doc_content, doc_metadata)

        # Add chunks to database
        db.add_documents(st_chunks)

        # Remove the temporary file after processing
        os.remove(temp_file_path)

def run_rag_chain(query):
    # Create a Retriever Object and apply Similarity Search
    retriever = db.as_retriever(search_type="similarity", search_kwargs={'k': 5})

    # Initialize a Chat Prompt Template
    PROMPT_TEMPLATE = """
    You are a highly knowledgeable assistant specializing in pharmaceutical sciences. 
    Answer the question based only on the following context:
    {context}

    Answer the question based on the above context:
    {question}

    Use the provided context to answer the user's question accurately and concisely.
    Don't justify your answers.
    Don't give information not mentioned in the CONTEXT INFORMATION.
    Do not say "according to the context" or "mentioned in the context" or similar.
    """

    prompt_template = ChatPromptTemplate.from_template(PROMPT_TEMPLATE)

    # Initialize a Generator (i.e. Chat Model)
    chat_model = ChatGoogleGenerativeAI(
        model="gemini-1.5-pro",
        api_key=st.session_state.get("gemini_api_key"),
        temperature=1
    )

    # Initialize a Output Parser
    output_parser = StrOutputParser()

    # RAG Chain
    rag_chain = {"context": retriever | format_docs, "question": RunnablePassthrough()} | prompt_template | chat_model | output_parser

    # Invoke the Chain
    response = rag_chain.invoke(query)

    return response

def main():
    st.set_page_config(page_title="PharmaQuery", page_icon=":microscope:")
    st.header("Pharmaceutical Insight Retrieval System")

    query = st.text_area(
        ":bulb: Enter your query about the Pharmaceutical Industry:",
        placeholder="e.g., What are the AI applications in drug discovery?"
    )

    if st.button("Submit"):
        if not query:
            st.warning("Please ask a question")
        
        else:
            with st.spinner("Thinking..."):
                result = run_rag_chain(query=query)
                st.write(result)

    with st.sidebar:
        st.title("API Keys")
        gemini_api_key = st.text_input("Enter your Gemini API key:", type="password")

        if st.button("Enter"):
            if gemini_api_key:
                st.session_state.gemini_api_key = gemini_api_key
                st.success("API key saved!")

            else:
                st.warning("Please enter your Gemini API key to proceed.")
    
    with st.sidebar:
        st.markdown("---")
        pdf_docs = st.file_uploader("Upload your research documents related to Pharmaceutical Sciences (Optional) :memo:",
                                    type=["pdf"],
                                    accept_multiple_files=True
        )
        
        if st.button("Submit & Process"):
            if not pdf_docs:
                st.warning("Please upload the file")

            else:
                with st.spinner("Processing your documents..."):
                    add_to_db(pdf_docs)
                    st.success(":file_folder: Documents successfully added to the database!")

    # Sidebar Footer
    st.sidebar.write("Built with ❤️ by [Charan](https://www.linkedin.com/in/codewithcharan/)")
             
if __name__ == "__main__":
    main()