app.py

import streamlit as st  
from langchain_core.messages import HumanMessage, AIMessage, SystemMessage
from langchain_core.prompts import ChatPromptTemplate, SystemMessagePromptTemplate, HumanMessagePromptTemplate
import os 
import nltk 
nltk.download("punkt")

st.title(':blue[Langchain:] A Rag System on “Leave No Context Behind” Paper')
st.header("AI Chatbot :robot_face:")

os.environ["GOOGLE_API_KEY"] = os.getenv("k4")
# Creating a template

chat_template = ChatPromptTemplate.from_messages([
    # System Message establishes bot's role and general behavior guidelines
    SystemMessage(content="""You are a Helpful AI Bot. 
    You take the context and question from user. Your answer should be based on the specific context."""),
    # Human Message Prompt Template
    HumanMessagePromptTemplate.from_template("""Answer the question based on the given context.
    Context:
    {context} 
    
    Question: 
    {question}
    
    Answer: """)
])

#user's question.
#how many results we want to print.

from langchain_google_genai import ChatGoogleGenerativeAI  

chat_model = ChatGoogleGenerativeAI(model="gemini-1.5-pro-latest")

from langchain_core.output_parsers import StrOutputParser

output_parser = StrOutputParser()

chain = chat_template | chat_model | output_parser

from langchain_community.document_loaders import PDFMinerLoader
from langchain_text_splitters import NLTKTextSplitter

uploaded_file = st.file_uploader("Choose a pdf file",type = "pdf")

pdf_loader = PDFMinerLoader(uploaded_file)
dat_nik = pdf_loader.load()
text_splitter = NLTKTextSplitter(chunk_size = 500,chunk_overlap = 100)
chunks = test_splitter.split_documents(dat_nik)
    
# dat = PDFMinerLoader("2404.07143.pdf")
# dat_nik =dat.load()
# # Split the document into chunks


# text_splitter = NLTKTextSplitter(chunk_size=500, chunk_overlap=100)

# chunks = text_splitter.split_documents(dat_nik)
# Creating Chunks Embedding
# We are just loading OpenAIEmbeddings
from langchain_google_genai import GoogleGenerativeAIEmbeddings  

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

# vectors = embeddings.embed_documents(chunks)
# Store the chunks in vector store
from langchain_community.vectorstores import Chroma  

# Creating a New Chroma Database
db = Chroma.from_documents(chunks, embedding_model, persist_directory="./chroma_db_1")

# saving the database on drive
db.persist()
# Setting a Connection with the ChromaDB
db_connection = Chroma(persist_directory="./chroma_db_", embedding_function=embedding_model)
# Converting CHROMA db_connection to Retriever Object, which retrieves top 5 results
retriever = db_connection.as_retriever(search_kwargs={"k": 5})


from langchain_core.runnables import RunnablePassthrough #takes user's question.

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

# format chunks: takes the 5 results, combines all the chunks and displays one output.
rag_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | chat_template
    | chat_model
    | output_parser
)

user_input = st.text_area("Ask Questions to AI")
if st.button("Submit"):
    st.subheader(":green[Query:]")
    st.subheader(user_input)
    response = rag_chain.invoke(user_input)
    st.subheader(":green[Response:-]")
    st.write(response)