openvinotoolkit/openvino ? reverse-engineered prompt

Build me an open source toolkit for making trained AI models run faster and easier in real apps. I want someone to take a model they already made in PyTorch, TensorFlow, ONNX, Keras, PaddlePaddle, or JAX, convert it, optimize it, and run inference without needing the original training framework installed. It should work for common use cases like image recognition, object detection, speech recognition, text models, recommendation models, and generative AI.

Make it usable from Python and C++ first, with clear examples that show loading a model, converting it, choosing CPU, Intel GPU, or Intel NPU, and running predictions. Include sample projects, tests, build scripts, and friendly documentation so developers can install it quickly and learn from tutorials. Keep the focus on fast inference from edge devices to cloud machines, not training models. Look up current docs online if you need to.