Deep Learning Model Orchestration Framework

I need a concise yet extensible orchestration framework focused exclusively on deep-learning workflows. The core requirement is the ability to evaluate and then deploy models built in TensorFlow or PyTorch without touching training logic. Your framework should: • Accept trained TensorFlow or PyTorch artefacts, run automated evaluation against a configurable test set, summarise the metrics, and log everything in an easily queryable format. • A judge model should select a model among a set of open source and commercial offering only when it passes the evaluation threshold that is defined, then expose it through a predictable deployment interface (REST or gRPC is fine if clearly documented). I expect clean, well-structured code, a short README that shows how to plug in a new model, and a demo script that walks from artefact upload through to live endpoint. I will consider the job complete once I can reproduce the flow on my own machine and see evaluation results followed by a working deployment.