Build a lightweight 'model-zoo' management tool that automatically optimizes models (quantization/pruning) based on the specific hardware detected during deployment.
Suggested repo: nano-tune
"Train once, optimize for every device automatically."
Estimated effort: 70h