I have a growing archive of high-resolution intraoral photos and I need an AI mechanism that can automatically locate every tooth in each image, assign the correct FDI or Universal numbering, and clearly classify each tooth type (incisor, canine, premolar, molar, wisdom). You are free to decide whether a pre-trained or custom-built model is best; accuracy and speed matter more to me than the underlying brand of framework. Python with PyTorch or TensorFlow, OpenCV for preprocessing, and a clean REST or CLI inference interface would suit my workflow, but if you have a more elegant stack, I’m open. Key deliverables • A trained model capable of processing new intraoral 3d scans and returning JSON with tooth positions, numbers and classes. • Lightweight script or API endpoint to run the model locally or on a small cloud instance. • Brief documentation covering installation, expected input format, output schema and a quick test example. Acceptance criteria 1. ≥ 95 % tooth-level detection recall on my validation set. 2. Correct numbering and type classification on ≥ 93 % of detected teeth. 3. End-to-end inference time ≤ 2 s per image on a mid-range GPU. Once these benchmarks are hit, I’m ready to integrate the model into my chairside software, so clear, maintainable code is a must. Let me know your proposed approach and any data needs you foresee.