Style Transfer Project Using Computer Vision

I'm looking for someone to build and document a neural style transfer system in Python using TensorFlow and Keras, based on the optimisation method from Gatys et al. (2016). The core is a working pipeline that takes a content photo and a style painting and produces a stylised image using a pre-trained VGG network as a fixed feature extractor, with the content loss, Gram-matrix style loss and total variation loss implemented from scratch rather than pulled from a library wrapper. Beyond a single working run, I need the pipeline wrapped in an experimental harness that varies the backbone (VGG-16 against VGG-19), varies the style-content weighting, and records loss values and output images for each configuration so results can be compared side by side and reproduced. Deliverables are the annotated Colab notebook, a gallery of style-content combinations, and a written technical report covering the literature, the design rationale, the evaluation method and the results. Evaluation should combine recorded loss figures with a small structured viewer survey, since loss alone doesn't capture output quality. Whoever takes this on needs to understand what the layer choices and weightings actually do, not just run a tutorial to completion.