server/inference.py

from langchain.schema.runnable import RunnablePassthrough
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_core.runnables import RunnableLambda

from llm.gemini import questions_template, format_questions_instructions, questions_parser
from data.load_data import retriever

def get_questions(_dict):
    question = _dict["question"]
    context = _dict["context"]
    messages = questions_template.format_messages(
        context=context,
        question=question,
        format_questions_instructions=format_questions_instructions,
    )

    chat = ChatGoogleGenerativeAI(model="gemini-pro")
    response = chat.invoke(messages)
    return questions_parser.parse(response.content)


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


rag_chain = {
    "context": retriever | RunnableLambda(format_docs),
    "question": RunnablePassthrough(),
} | RunnableLambda(get_questions)