modules/quiz/quiz_helper.py

import random
from typing import Literal
from config import SanatanConfig
from modules.quiz.models import Question
from sanatan_assistant import query, allowedCollections
from openai import OpenAI

client = OpenAI()


def generate_question(
    collection: allowedCollections,
    complexity: Literal["beginner", "intermediate", "advanced"],
    mode: Literal["mcq", "open"],
    preferred_lamguage: str = "English",
) -> Question:
    """
    Fetch a random scripture record and have the LLM generate a structured Question.
    """
    print("Generating question ...", collection, complexity,mode, preferred_lamguage)
    # 1. Fetch random scripture record
    context = query(
        collection_name=collection,
        query=None,
        metadata_where_clause=None,
        n_results=1,
        search_type="random",
    )
    if not context:
        raise ValueError(f"No records found in collection {collection}")

    # 2. Prompt (grounded in record only)
    prompt = f"""
    You are a quiz generator. Use ONLY the following scripture record to create a question.

    Context from {collection}:
    {context}

    Rules:
    - Do not invent facts beyond the context.
    - Difficulty level: {complexity}
    - Mode: {mode}
    - If mode is 'mcq', generate 3–4 plausible choices (with one correct).
    - If mode is 'open', leave 'choices' empty and provide a reference answer.
    - Provide all fields in JSON. The `native_lyrics` field MUST always be populated from the context, even if short. Do not omit.
    - Ensure all text fields, including 'native_lyrics', are valid JSON strings.
    - Escape all newlines as \\n. Do not omit any part of the verse.    
    - User's preferred language is {preferred_lamguage}. Translate everything except the native verses to this language.
    """

    # 3. Structured response with Pydantic class reference
    response = client.chat.completions.parse(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
        response_format=Question,
    )

    # print(response)

    return response.choices[0].message.parsed


# Example usage
if __name__ == "__main__":
    for i in range(3):
        q = generate_question(
            collection=random.choice(
                [
                    s["collection_name"]
                    for s in SanatanConfig.scriptures
                    if s["collection_name"] != "yt_metadata"
                ]
            ),
            complexity=random.choice(["beginner", "intermediate", "advanced"]),
            mode=random.choice(["mcq", "open"]),
            preferred_lamguage="Tamil",
        )
        print(q.model_dump_json(indent=1))
        print("_______________________")