AI Product Image Classification App

I need an end-to-end application that can automatically classify product images into their correct categories. The input will always be standard product photos; no medical or satellite data is involved. Your job is to design, train, validate, and package an image-classification model that can recognise each product class with high accuracy.

You’re free to choose the framework you prefer—TensorFlow, PyTorch, or a comparable deep-learning library—as long as the final solution is easy for my in-house team to maintain. A well-structured Python codebase, clear preprocessing steps, and reproducible training scripts are essential. If you have a pre-trained backbone such as ResNet, EfficientNet, or MobileNet that speeds up training, feel free to leverage it; just ensure you fine-tune correctly and document every dependency.

I will provide a labelled dataset plus a short list of target categories. Please include in your proposal how you plan to handle class imbalance, data augmentation, and overfitting. The finished system should expose a simple REST or CLI interface where we can pass a folder or single image and receive the predicted class name with a confidence score.

Deliverables:
• Clean, well-commented source code and requirements.txt/conda-environment.yml
• Trained model weights and the training notebook or script
• Inference module or microservice (Flask/FastAPI or similar)
• Brief README covering setup, usage, and retraining steps
• Performance report: accuracy, precision/recall, confusion matrix, and a short explanation of misclassifications

Acceptance criteria:
• ≥90 % overall accuracy on a held-out validation set we will share after project kick-off
• Inference time ≤300 ms per 224×224 image on a single GPU or CPU fallback

If this fits your skill set, let me know your preferred tech stack and estimated timeline.