config.py

from pathlib import Path
import os

# ── Load .env if python-dotenv available ──────────────────────────────────────
try:
    from dotenv import load_dotenv
    load_dotenv(Path(__file__).parent / ".env", override=False)
except ImportError:
    pass  # dotenv optional — env vars can be set by OS / Docker

# ── Directories ────────────────────────────────────────────────────────────────
PROJECT_ROOT  = Path(__file__).parent.resolve()   # relative — works on any machine
DATA_DIR      = PROJECT_ROOT / "data"
INDEX_DIR     = PROJECT_ROOT / "index"
LOGS_DIR      = PROJECT_ROOT / "logs"

# ── Source data (KCC parquet files) ───────────────────────────────────────────
KCC_DATA_DIR  = Path(os.environ.get(
    "KCC_DATA_DIR",
    str(PROJECT_ROOT / "data" / "kcc_raw")
))

# ── Step 1 output ──────────────────────────────────────────────────────────────
CLEAN_DATA_DIR  = DATA_DIR / "clean_years"
CLEAN_DATA_FILE = DATA_DIR / "kcc_clean.parquet"

# ── Step 2 output ──────────────────────────────────────────────────────────────
FAISS_INDEX_FILE   = INDEX_DIR / "faiss.index"
METADATA_FILE      = INDEX_DIR / "metadata.parquet"
BM25_INDEX_FILE    = INDEX_DIR / "bm25.db"

# ── Embedding model ────────────────────────────────────────────────────────────
EMBEDDING_MODEL  = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
EMBEDDING_DIM    = 384
EMBED_BATCH_SIZE = 512

# ── Retrieval settings ─────────────────────────────────────────────────────────
TOP_K = 5

# ── Cross-encoder reranker ─────────────────────────────────────────────────────
RERANKER_MODEL   = "cross-encoder/mmarco-mMiniLMv2-L12-H384-v1"
RERANKER_TOP_N   = 20
MIN_ANSWER_CHARS = 100

# ── LLM cascade — keys from environment ───────────────────────────────────────
GROQ_API_KEY        = os.environ.get("GROQ_API_KEY", "")
GROQ_MODEL_PRIMARY  = "meta-llama/llama-4-scout-17b-16e-instruct"
GROQ_MODEL_FALLBACK = "llama-3.3-70b-versatile"

GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY", "")
GEMINI_MODEL   = "gemma-3-27b-it"

# ── Mandi price API ───────────────────────────────────────────────────────────
DATA_GOV_API_KEY = os.environ.get("DATA_GOV_API_KEY", "")

# ── Environment ───────────────────────────────────────────────────────────────
ENVIRONMENT    = os.environ.get("ENVIRONMENT", "development")
LOG_LEVEL      = os.environ.get("LOG_LEVEL", "INFO")
ALLOWED_ORIGINS = [
    o.strip() for o in
    os.environ.get("ALLOWED_ORIGINS", "http://localhost:8501,http://localhost:8000").split(",")
    if o.strip()
]
B2B_DEMO_PASSWORD = os.environ.get("B2B_DEMO_PASSWORD", "DeHaat2025")

# ── Data filters ───────────────────────────────────────────────────────────────
MIN_QUERY_LEN  = 10
MIN_ANSWER_LEN = 20
START_YEAR      = 2006
CHUNK_SIZE      = 100000
SAMPLE_PER_YEAR = None

# ── Column names ───────────────────────────────────────────────────────────────
QUERY_COL  = "QueryText"
ANSWER_COL = "KccAns"

# Create dirs on import
for _d in [DATA_DIR, CLEAN_DATA_DIR, INDEX_DIR, LOGS_DIR]:
    _d.mkdir(parents=True, exist_ok=True)