Show HN: 16-year-old built a NASA-grade AI ensemble detecting real exoplanets (github.com)

A 16-year-old developer released "The Council of Lords" (aka MAVS), an AI ensemble that claims NASA-grade exoplanet detection on real Kepler and TESS data. The system combines five specialized neural networks plus a four-method period detector (Box Least Squares, autocorrelation, transit-timing, Lomb–Scargle) whose weighted consensus and phase‑fold stacking boost SNR (√N stacking) to recover extremely weak transits (they report detecting a 0.3% depth buried in 0.8% noise). MAVS integrates TIC/Gaia DR3/KIC stellar parameters, applies physics-based filters (Kepler’s 3rd law, Stefan–Boltzmann equilibrium temps), corrects BLS/phase-folding bugs, increases period resolution to 0.01 days, and exposes per-model votes, red‑flagging for instrument systematics. The project also includes 2D orbit visualization, habitability scoring, and realtime frontend physics validation with claimed millisecond response. Significance: the approach illustrates how multi-method ensembles plus domain-aware physics constraints can tighten vetting and reduce false positives, potentially accelerating candidate triage for observatories and serving educational use. Caveats: the extraordinary 100% accuracy claims (36/36 confirmed, perfect on “brutal” tests) require independent validation—risk of dataset leakage, overfitting, or tuned test selection remains. If reproducible on blind, diverse survey data and under different systematics, this paradigm could materially improve automated vetting; until then, the community should review code, datasets, and evaluation protocols.