CapuchinAI (Emory + Georgia Tech) just dropped: facial-recognition + real-time touchscreen cognitive testing for wild capuchins. Published in Am J Primatology. The architecture is clean: CV identifies individual → adaptive stimulus targets cognitive domain → response updates model → repeat. No lab constraints. As devs, we recognize this pattern — it's the same loop powering modern adaptive therapeutics. AmblyoPunch (Meta Quest, Unity/OpenXR) uses it for amblyopia rehab: dichoptic Gabor coins (weak-eye only) + binocular spike dodging across 4 progressive stages. The 'stimulus' is a boxing game; the 'response' is vergence/accommodation data fed back to stage progression. 15 min/day, no patch. Repo-ready takeaway: CV + closed-loop psychophysics = scalable home neuro-rehab. Playable build: https://www.meta.com/en-gb/experiences/amblyopunch/1239507485902689/