Let’s build an intelligent future.

Satellite fleet health

A health check for a fleet of satellites: AI spots which parts are wearing out and says what to do about it.

Physically-grounded synthetic telemetry, per-subsystem failure models, and an operations dashboard that turns predictions into recommended actions — including a physics-driven end-of-life deorbit call. The satellites are simulated; the physics and the ML are not.

• Comms-fault detection rebuilt from F1 0.14 to 0.89 by diagnosing a threshold bug
• Per-subsystem decision thresholds tuned by cross-validation, not a default 0.5 cutoff
• Orbital decay and station-keeping burns modeled from real drag physics

Teaching AI to predict how many more flights a jet engine has before it needs attention.

An attention-based LSTM feeding an XGBoost regressor predicts how many cycles a turbofan engine has left, trained on NASA's C-MAPSS run-to-failure data. Benchmarked against a baseline LSTM under identical preprocessing and early stopping.

• Attention highlights the degradation-relevant timesteps in each 50-cycle window
• Learned context vector fused with window statistics before gradient boosting
• Evaluated on FD001 and on FD004, with six operating regimes and two fault modes