Machine learning enables high-throughput, spectrometer-free chiral discrimination via circularly-polarized dark-field imaging.

Chiral discrimination is crucial for biochemical applications, yet conventional spectroscopic methods suffer from low throughput due to serial acquisition. Here, we present a high-throughput strategy using machine learning on left- and right-circularly polarized dark-field RGB images, eliminating the need for spectral acquisition. Our approach achieves >97% accuracy for strongly chiral nanoparticles and generalizes to unseen nanoparticle types with 74% accuracy. As a proof-of-principle for molecular chirality discrimination, we detect chiral molecules adsorbed on plasmonic nanoparticles with 80% accuracy. While cross-molecule generalization remains limited as a consequence of molecule-specific optical signatures, it can be addressed via a library-based fingerprinting approach. This work establishes a scalable paradigm for high-throughput chirality discrimination with particular promise for pharmaceutical enantiomer screening.