config.py

from __future__ import annotations

import os
from pathlib import Path
from typing import Optional

from dotenv import load_dotenv
from loguru import logger

load_dotenv()

BASE_DIR = Path(__file__).resolve().parent

CLAIMS_DATA_ROOT = Path(os.getenv("CLAIMS_DATA_ROOT", BASE_DIR / "data" / "claims"))
SCHEDULING_DATA_ROOT = Path(
    os.getenv("SCHEDULING_DATA_ROOT", BASE_DIR / "data" / "scheduling")
)

CLAIM_PACKETS_ROOT = CLAIMS_DATA_ROOT / "mock_claim_packets"
CLAIMS_POLICY_RAG_JSONL = (
    CLAIMS_DATA_ROOT / "rag" / "policy_benefit_rag" / "policy_benefit_chunks.jsonl"
)
CLAIMS_EXCEPTION_RAG_JSONL = (
    CLAIMS_DATA_ROOT
    / "rag"
    / "exception_similarity_rag"
    / "resolved_exception_cases.jsonl"
)

SCHEDULING_MOCK_DATA_ROOT = SCHEDULING_DATA_ROOT / "mock_data"
SCHEDULING_PROVIDER_RAG_JSONL = (
    SCHEDULING_DATA_ROOT
    / "rag"
    / "provider_specialty_matching"
    / "provider_specialty_profiles.jsonl"
)

OPENAI_CHAT_MODEL = os.getenv("OPENAI_CHAT_MODEL", "gpt-4o-mini")
OPENAI_EMBED_MODEL = os.getenv("OPENAI_EMBED_MODEL", "text-embedding-3-small")
HF_FT_EMBED_MODEL_URL = os.getenv("HF_FT_EMBED_MODEL_URL")

# Cost controls. Keep paid calls explicit for demos/HF Spaces.
USE_LLM = os.getenv("USE_LLM", "false").lower() == "true"
USE_PAID_EMBEDDINGS = os.getenv("USE_PAID_EMBEDDINGS", "false").lower() == "true"

LOG_LEVEL = os.getenv("LOG_LEVEL", "INFO")
logger.remove()
logger.add(lambda msg: print(msg, end=""), level=LOG_LEVEL)

logger.info("Config loaded")
logger.info(f"USE_LLM={USE_LLM}; USE_PAID_EMBEDDINGS={USE_PAID_EMBEDDINGS}")
logger.debug(f"CLAIMS_DATA_ROOT={CLAIMS_DATA_ROOT}")
logger.debug(f"SCHEDULING_DATA_ROOT={SCHEDULING_DATA_ROOT}")


def build_llm() -> Optional[object]:
    """Create the shared chat LLM only when explicitly enabled."""
    if not USE_LLM:
        logger.info("LLM disabled. Set USE_LLM=true to enable chat-completion calls.")
        return None

    try:
        from langchain_openai import ChatOpenAI

        logger.info(f"Creating ChatOpenAI model={OPENAI_CHAT_MODEL}")
        return ChatOpenAI(model=OPENAI_CHAT_MODEL, temperature=0)
    except Exception as exc:
        logger.warning(
            f"Could not create LLM. Falling back to deterministic logic. Error: {exc}"
        )
        return None


LLM = build_llm()