1012 Hill Street Planes

Overview

A self-hosted aircraft-tracking platform that decodes ADS-B radio traffic, records durable receiver archives, and renders live aircraft positions in a React map. Its learned flight predictor was evaluated against constant-velocity extrapolation on a chronologically held-out test set.

Key Features

• Live map of decoded aircraft positions with altitude-coded aircraft markers
• Durable radio-capture archives with receiver coverage and uptime reporting
• Observed 2,619 distinct aircraft addresses and 3,894,586 decoded packets on October 2, 2026
• Recorded a one-day peak of 130 decoded messages per second
• Reached a 469.8 km observed maximum accepted position range
• Learned predictor reduced held-out 60-second flight position error by 19.6% versus constant-velocity extrapolation

Technical Challenges

Combined live radio decoding, position validation, archival storage, map rendering, and offline model evaluation into a receiver that remains useful during restarts and intermittent connectivity. The forecast comparison keeps whole aircraft encounters within one chronological partition so the final test does not leak observations from the same flight.

Outcome

As of October 3, 2026, the receiver had accumulated 25.2 days of observing time. On the final held-out test, its 60-second model achieved 0.377 km mean position error across 298 flights, compared with 0.469 km for the constant-velocity baseline.

Technologies

• React
• ADS-B
• Software-Defined Radio
• Machine Learning
• NDJSON