I need an end-to-end OCR solution that can reliably read alphanumeric characters embossed on flat metal parts so they can be catalogued automatically in our inventory management system. The parts move under a fixed industrial camera, and the lighting is controlled but reflections are still a challenge, so the network must cope with glare, minor scratches, and variable contrast. Here is what I am looking for: • A Python-based pipeline that uses OpenCV for preprocessing (glare reduction, ROI extraction, perspective correction) and a deep-learning backbone—PyTorch or TensorFlow are both fine—trained specifically on metal-embossed text. • Model must detect and recognise single-line or multi-line embossing in real time (≥10 FPS on an NVIDIA T4, batch 1). • Accuracy target: ≥98 % character-level accuracy on a held-out test set we will provide, consisting solely of flat metal samples. • Clean, well-commented code plus an inference script that posts the recognised strings to a REST endpoint we will supply. • A brief training guide so we can extend the model later with new characters or alloys. Please include any similar computer-vision work you have done with reflective surfaces or embossed/engraved text and outline your proposed toolchain and timeline.