This work evaluates the effectiveness of the lightweight SqueezeNet CNN architecture for automated early detection of eye diseases from retinal scans on resource-constrained mobile devices, in order to guide the development of accurate and efficient screening systems deployable in low-resource settings.