Real historical outcomes
Trained on tens of thousands of self-reported decisions from public forums, structured into (features, outcome) pairs. Not scraped middle-50 percentiles.
§ Suite · Services
What Illuma actually does.
▸Live previewClick to collapse
Personal admit probability · your profile
ML · trained on real outcomes
reach
MIT
4%→11%
reach
Carnegie Mellon SCS
6%→14%
match
Georgia Tech CS
17%→34%
match
UIUC CS
32%→51%
safety
Purdue CS
45%→68%
Reported admit rate
Your personal probability
Output includes
- Free — no account required
- Admit-probability estimates across every top-30 US college
- ED / EA / REA round differentials, not one 'admit rate' averaged across rounds
- Hook adjustments (first-gen, URM, Pell, legacy) with a calibration step against real outcomes
- Impacted-major recognition — CS at Michigan is scored differently than the school-wide rate
Under the hood
ML models + calibration layers, not a ChatGPT wrapper.
Round-specific features
ED / EA / REA are separate features with school-specific boost calibration — Yale SCEA behaves differently from generic EA and the model knows.
Post-hoc calibration
Raw model probabilities are Platt-calibrated per admit-rate tier so a reported 22% actually means ~22%, not 'somewhere between 5% and 60%'.