A project studio where middle- and high-schoolers do genuine AI research on real aerospace data — training models on Kaggle experimental sets and NASA's CEA simulation code, then shipping a judge-ready Streamlit dashboard. Every stage runs in code and simulation.
Train ML models to predict solid-propellant burning rate from a real 137-sample Kaggle set, then generate thousands of simulated rows with NASA's CEA (rocketcea) and compare. That comparison is the original contribution.
Map elemental composition of organic waste to Higher Heating Value. Two on-ramps: no-code Orange Data Mining for younger students, graduating into scikit-learn. Strong ESG / climate framing.
Train a YOLOv8 crack-detection model on Roboflow's public surface-crack data, augmented with webcam photos of clay mock grains students make themselves. The live webcam demo is the showstopper.
Every track climbs the same ladder. Students start by reproducing a verified baseline in week 1, add one original study, publish it, and shape it into a competition submission.
Competition eligibility and rules change each year; participation routes are confirmed against official guidelines before any submission. See CIT's ISEF feeder-fair map and STEM competition guide.
We look at where the student is, then map which of the three tracks fits — all in code and data, nothing physical.