Research

Research and open-source work

My research sits where space systems meet machine learning. The interesting question is rarely whether a model can predict something — it’s how much anyone should trust the prediction when they have to act on it.

ORCID 0009-0008-8324-3847

Primary project

BEACON

Research prototype · Open source

BEACON

Bayesian Event Assessment for Conjunction Observation and Notification

A reproducible research project I created for calibrated, uncertainty-aware satellite conjunction triage using public CDM data — deciding which close approaches between objects on orbit warrant human attention. It evaluates rare-event ranking, probability calibration, uncertainty-aware human review, repeated split robustness, current-risk feature ablation, leakage-safe evaluation, and interactive visual analytics for model-grounded triage inspection.

Trustworthy AISatellite conjunction assessment Uncertainty quantificationCalibration Rare eventsBayesian ML Space-domain awareness

Research prototype only. BEACON is not an operational system and is not a collision-warning service. It is a reproducible research artifact for studying how calibrated machine learning behaves on public conjunction data.

Versions, DOIs, and licensing

Current development version

v0.3.0-rc1

Release candidate · no final version DOI yet

Latest archived artifact

v0.2.2

Reproducible release on Zenodo

Version DOI

10.5281/zenodo.21209794

Cites the exact archived v0.2.2 artifact

Concept DOI

10.5281/zenodo.21209119

Cites BEACON across all versions

Licensing

Two licenses apply, depending on which version you use

Important distinction Current main branch source: Apache License 2.0. Beginning with v0.3.0-rc1, BEACON source is Apache-2.0. · Archived v0.2.2 and earlier: MIT License. The Zenodo version DOI above points at an MIT-licensed artifact. · Documentation, figures, research materials, and demo scenarios: CC BY 4.0 unless otherwise noted.

Repository Version DOI Concept DOI

Companion project

ORACLE-Ω

A simulation-only research prototype for 3D spatial assurance visualization — an exploration of how spatial assurance problems can be represented and inspected in three dimensions. Code under Apache License 2.0; documentation under CC BY 4.0.

Simulation only. ORACLE-Ω is an early-stage research prototype with no operational use.

Repository

Interests

Where the work is heading

Trustworthy AI

Calibration, uncertainty quantification, and the question of when a model’s stated confidence can be relied on. A model that says ninety percent should be right ninety percent of the time.

Space systems

Orbital dynamics, space-domain awareness, and Earth observation — how objects and sensors behave on orbit, and what that constrains about anything inferred downstream.

Rare events & robustness

Ranking and evaluation when positives are scarce: repeated-split robustness, leakage-safe evaluation, and ablations that show which signal is actually doing the work.

Crowdsourced research

Over a thousand classifications contributed to research projects on Zooniverse, where volunteer analysis feeds directly into published science. Contributing since 2022.

Zooniverse profile