backend/utils/ocr.py

import os
import requests


HF_MODEL = os.getenv("HF_OCR_MODEL", "microsoft/trocr-base-printed")


def run_hf_ocr(image_bytes: bytes) -> str:
    token = os.getenv("HUGGINGFACE_API_TOKEN")
    if not token:
        raise RuntimeError("Missing HUGGINGFACE_API_TOKEN")

    # Use Inference API for text recognition
    url = f"https://api-inference.huggingface.co/models/{HF_MODEL}"
    headers = {"Authorization": f"Bearer {token}"}
    response = requests.post(url, headers=headers, data=image_bytes, timeout=60)
    if response.status_code >= 400:
        raise RuntimeError(f"HF API error: {response.status_code} {response.text}")

    try:
        data = response.json()
    except Exception as exc:
        raise RuntimeError(f"Invalid HF response: {exc}")

    # trocr returns dicts with 'generated_text'
    if isinstance(data, list) and data and isinstance(data[0], dict):
        text = data[0].get("generated_text")
        if isinstance(text, str):
            return text

    # fallback to raw
    return str(data)