API Development for Apple Grading AI

We have a trained, working YOLO-based object detection + classification model (Python) that detects and grades apples in tray images as Grade A / Grade B
• Need it wrapped into a callable REST API (FastAPI or Flask preferred) that accepts an image and returns grading results
• Must be optimized for GPU inference (RunPod, RTX A5000) — not CPU-only
• Model loading must be efficient (loaded once, kept warm — not reloaded per request)
• Needs to handle real-world phone-camera photos, not just clean training images — proper image preprocessing/normalization required
• Robust error handling: bad images, no apples detected, corrupted uploads, timeouts
• Must support multiple concurrent requests without performance degradation (10+ concurrent users expected)
• Deployment included — API should be live and testable, not just handed over as code