import os

from dotenv import load_dotenv

from pinecone import Pinecone

from langchain_openai import OpenAIEmbeddings, ChatOpenAI

from langchain_pinecone import PineconeVectorStore

from langchain_community.document_loaders import PyPDFDirectoryLoader

from langchain_text_splitters import RecursiveCharacterTextSplitter

from langchain_core.prompts import ChatPromptTemplate

from langchain_core.runnables import RunnablePassthrough

from langchain_core.output_parsers import StrOutputParser

# 1. Configuration: Load environment variables from .env

load_dotenv()

# 2. Initialization: Connect to the Pinecone vector database

pc = Pinecone(api_key=os.getenv("PINECONE_API_KEY"))

index_name = "sentry-index"

# 3. Ingestion: Load PDF files from the local 'docs' directory

loader = PyPDFDirectoryLoader("./docs")

docs = loader.load()

# 4. Processing: Split documents into chunks for retrieval

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)

splits = text_splitter.split_documents(docs)

# 5. Vectorization: Generate 1536-dimension embeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-small")

vectorstore = PineconeVectorStore.from_documents(

splits, embeddings, index_name=index_name

)

# 6. Pipeline: Define the RAG logic using LCEL (Modern Syntax)

llm = ChatOpenAI(model="gpt-4o", temperature=0)

retriever = vectorstore.as_retriever()

template = """Answer the question based only on the following context:

{context}

Question: {question}

"""

prompt = ChatPromptTemplate.from_template(template)

# This chain bypasses the 'langchain.chains' module entirely

rag_chain = (

{"context": retriever, "question": RunnablePassthrough()}

| prompt

| llm

| StrOutputParser()

)

# 7. Execution: Execute query and output result

if __name__ == "__main__":

#query = "What are the key technical requirements mentioned in this document?"

query = "Summarize the main purpose of this document."

response = rag_chain.invoke(query)

print("\n--- SentryQuery Result ---")

print(response)