run_batch.py

"""
Runs the MenuAI pipeline against every photo in a folder, one at a time,
and saves all results to a single JSON file for later review/labeling.

Usage:
    python run_batch.py path/to/photos_folder
"""
from __future__ import annotations

import json
import sys
import time
from pathlib import Path

from menuai_pipeline import generate_menu_entry

# Free-tier APIs limit how many requests you can make per minute.
# Waiting a couple seconds between photos avoids hitting that limit.
SECONDS_BETWEEN_PHOTOS = 3

# Which file extensions count as a photo we should process
IMAGE_EXTENSIONS = {".jpg", ".jpeg", ".png", ".webp"}


def find_photos(folder: Path) -> list[Path]:
    """Return every image file directly inside the given folder, sorted by name."""
    return sorted(
        p for p in folder.iterdir()
        if p.is_file() and p.suffix.lower() in IMAGE_EXTENSIONS
    )


def run_batch(folder_path: str, output_path: str = "batch_results.json") -> None:
    folder = Path(folder_path)
    photos = find_photos(folder)

    if not photos:
        print(f"No photos found in {folder}. Looking for: {IMAGE_EXTENSIONS}")
        return

    print(f"Found {len(photos)} photos. Starting batch run...\n")

    results = []
    for i, photo_path in enumerate(photos, start=1):
        print(f"[{i}/{len(photos)}] Processing {photo_path.name}...", end=" ", flush=True)

        try:
            comparison = generate_menu_entry(str(photo_path))
            results.append({
                "filename": photo_path.name,
                "result": json.loads(comparison.model_dump_json()),
            })
            status = "NEEDS REVIEW" if comparison.needs_review else "auto-approved"
            print(f"done — agreement {comparison.agreement_score:.2f} ({status})")

        except Exception as e:
            # One bad photo (corrupted file, API error, etc.) should not stop the batch.
            print(f"FAILED — {e}")
            results.append({
                "filename": photo_path.name,
                "result": None,
                "error": str(e),
            })

        # Don't sleep after the very last photo — no point waiting once we're done
        if i < len(photos):
            time.sleep(SECONDS_BETWEEN_PHOTOS)

    with open(output_path, "w", encoding="utf-8") as f:
        json.dump(results, f, indent=2, ensure_ascii=False)

    # Quick summary so you don't have to open the JSON file to see how it went
    succeeded = [r for r in results if r["result"] is not None]
    needs_review_count = sum(
        1 for r in succeeded if r["result"]["needs_review"]
    )
    print(f"\nDone. {len(succeeded)}/{len(photos)} photos processed successfully.")
    print(f"{needs_review_count} flagged for review, {len(succeeded) - needs_review_count} auto-approved.")
    print(f"Full results saved to {output_path}")


if __name__ == "__main__":
    if len(sys.argv) != 2:
        print("Usage: python run_batch.py path/to/photos_folder")
        sys.exit(1)

    run_batch(sys.argv[1])