langchain/templates/rag-opensearch/rag_opensearch/chain.py

import os

from langchain_community.chat_models import ChatOpenAI
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_community.vectorstores.opensearch_vector_search import (
    OpenSearchVectorSearch,
)
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.pydantic_v1 import BaseModel
from langchain_core.runnables import RunnableParallel, RunnablePassthrough

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
OPENSEARCH_URL = os.getenv("OPENSEARCH_URL", "https://localhost:9200")
OPENSEARCH_USERNAME = os.getenv("OPENSEARCH_USERNAME", "admin")
OPENSEARCH_PASSWORD = os.getenv("OPENSEARCH_PASSWORD", "admin")
OPENSEARCH_INDEX_NAME = os.getenv("OPENSEARCH_INDEX_NAME", "langchain-test")


embedding_function = OpenAIEmbeddings(openai_api_key=OPENAI_API_KEY)

vector_store = OpenSearchVectorSearch(
    opensearch_url=OPENSEARCH_URL,
    http_auth=(OPENSEARCH_USERNAME, OPENSEARCH_PASSWORD),
    index_name=OPENSEARCH_INDEX_NAME,
    embedding_function=embedding_function,
    verify_certs=False,
)


retriever = vector_store.as_retriever()


def format_docs(docs):
    return "\n\n".join([d.page_content for d in docs])


# RAG prompt
template = """Answer the question based only on the following context:
{context}
Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)

# RAG
model = ChatOpenAI(openai_api_key=OPENAI_API_KEY)
chain = (
    RunnableParallel(
        {"context": retriever | format_docs, "question": RunnablePassthrough()}
    )
    | prompt
    | model
    | StrOutputParser()
)


# Add typing for input
class Question(BaseModel):
    __root__: str


chain = chain.with_types(input_type=Question)