Engineering.
The code under the papers.
The MOSOF / NDCI work shouldn't have to be taken on trust. Three open-source artefacts package the published methods as runnable Python you can clone, read, and modify — plus an annotated walkthrough of the inner optimisation loop.
Reference implementations against the paper notation; not the production research code. Each repo cites the underlying paper and runs in seconds on a laptop. MIT-licensed.
Three runnable repos.
MOSOF reference implementation
NSGA-II driver over a binary sensor-selection space with three competing objectives — NDCI-weighted performance, suite cost, series-MTBF reliability. Produces a Pareto front and a knee solution in ~5 seconds.
NDCI calculator
Per-sensor Normalised Diagnostic Contribution Index over the published 12-fault × 18-sensor matrix. Re-ranks sensors under uniform, ECS-heavy, and Engine-heavy fault priors — the stakeholder-aware axis of MOSOF.
Pareto data pipeline
Loads the canonical B737-class cross-subsystem Pareto front (77 suites), verifies non-dominance, identifies the published knee, and emits clean CSV + a quick-look plot. The source dataset is /data/pareto-b737-ecs.json.
What "reference implementation" means here
These repos package the published method, not the original research code. The math (NSGA-II selection, NDCI prior-weighting, series-MTBF reliability, knee detection) follows the Sensors (MDPI) papers verbatim; the implementation is a fresh write-up against current Python tooling (pymoo, NumPy, Matplotlib) so the code is short enough to read in one sitting. Each repo cites the underlying paper, and the production research code lives with the thesis.
The inner loop, annotated.
One long-form essay annotating the multi-objective selection inner loop end-to-end — the GA step, the three-axis fitness function, the Pareto-front extraction, and the knee selection. Reads cleanly without the papers in hand.
MOSOF in ~100 lines of Python.
An annotated NSGA-II walkthrough that produces a publication-grade Pareto front for a synthetic-but-realistic sensor-selection problem. Every line traceable to the 2026 Sensors paper.
Read the walkthrough →Why this surface exists.
Hiring evaluators at AI / ML scale-ups and aerospace teams reasonably expect to see code before a screening call. The papers are the canonical record; the proof page is the case-study version. This page is the engineering surface — runnable, readable, MIT-licensed.
If you'd like the full thesis appendix code, or the production research scripts that built the published Pareto fronts, the route is [email protected].