GOOD: a self-powered single-sensor node — the ΔT itself powers one tiny reading + one spike. A real, useful, honest learning artifact.
NOT: viable as city cooling — Peltier COP is high only at a tiny lift; push real cooling and it crashes below 1 (burns more grid power than heat removed). Or as compute infrastructure (mW-scale gen, ~2–6% efficient).
For cooling a city → PV panels + hydroponics/evaporative + shade. Thermoelectrics stay a niche.