scripts/build_index.py

"""One-time ChromaDB indexing of the item catalog.

Runs SBERT (all-MiniLM-L6-v2) over every item's title + description and
stores the embedding in ChromaDB under data/chroma/. The collection is
persistent — subsequent runs skip indexing if it already exists.

After this runs, Task B's retrieval layer can do semantic search:
    "thoughtful literary fiction about identity" → 50 nearest items

Usage:
    python -m scripts.build_index
    python -m scripts.build_index --force   # rebuild from scratch
"""
from __future__ import annotations

import argparse
import logging

import pandas as pd

from core.config import settings
from core.retrieval import ItemRetriever

logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
log = logging.getLogger(__name__)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--force", action="store_true",
                    help="Rebuild the index from scratch (default: skip if already built)")
    args = ap.parse_args()

    items_path = settings.processed_dir / "items.parquet"
    if not items_path.exists():
        raise SystemExit(f"Items file not found at {items_path}. Run `python data/prepare_data.py` first.")

    items = pd.read_parquet(items_path)
    log.info(f"Loaded {len(items):,} items")

    retriever = ItemRetriever()
    log.info(f"Building index at {retriever.persist_dir}")
    retriever.build_index(items, batch_size=256, force=args.force)

    # Quick smoke test
    log.info("Smoke testing retrieval...")
    results = retriever.retrieve("thoughtful literary fiction about identity", k=5)
    log.info(f"Test query returned {len(results)} results:")
    for r in results:
        log.info(f"  [{r.distance:.3f}] {r.domain} | {r.title[:80]}")


if __name__ == "__main__":
    main()