# chain_recommendations.py
import json
from typing import Dict
from langchain import PromptTemplate, LLMChain
from models import chat_model

improved_recommend_prompt_template = PromptTemplate(
    input_variables=["problems"],
    template=(
        "You are a wellness recommendation assistant. Given the following problem severity percentages:\n"
        "{problems}\n\n"
        "Carefully analyze these percentages and consider nuanced differences between the areas. "
        "Your goal is to recommend the most appropriate wellness packages based on a detailed assessment of these numbers, "
        "not just fixed thresholds. Consider the following guidelines:\n\n"
        "- If one area is extremely high (above 70) while others are lower, prioritize a package targeting that area.\n"
        "- If multiple areas are high or near high (e.g., above 60), consider recommending multiple specialized packages or a comprehensive program.\n"
        "- If all areas are moderate (between 30 and 70), recommend a balanced wellness package that addresses overall health.\n"
        "- If all areas are low, a general wellness package might be sufficient.\n"
        "- Consider borderline cases and recommend packages that address both current issues and preventive measures.\n\n"
        "Return the recommended wellness packages in a JSON array format."
    )
)

# Initialize the improved recommendation chain
recommend_chain = LLMChain(llm=chat_model, prompt=improved_recommend_prompt_template)

def generate_recommendations(problems: Dict[str, float]) -> str:
    recommendations = recommend_chain.run(problems=json.dumps(problems))
    return recommendations.strip()