labounad/spinhance — reverse-engineered prompt

Build me a Python project called SpinHance that can take low field 90 MHz proton NMR spectra and predict the underlying spin system, including chemical shifts, scalar couplings, and proton counts, so the same molecule can be reproduced or simulated at other field strengths.

I want the full workflow to be usable from the command line. It should be able to screen SMILES into molecules with 8 proton spin groups, turn those molecules into spin system JSON records, simulate 90 MHz and 600 MHz spectra, train a deep learning model on the spectra and spin system data, and save checkpoints. Include a simple interactive dashboard where I can load a checkpoint, inspect one molecule at a time, compare predicted versus true shift and coupling matrices, overlay spectra, and run batch metrics like shift error, coupling error, and degeneracy accuracy.

Please make it practical to run locally, but also include scripts or guidance for training on a cloud GPU with S3 storage. Look up current docs online if you need to.