The inference engine focuses on the inference stage and aims to provide a platform-agnostic execution of the inference pipeline. To achieve this, operator fusion and optimized kernels are employed.