Strategy
Read the case
The argument, the evidence, and ten ranked project ideas you can re-weight yourself.
Software Opportunities Briefing
The funding context, the AI-weather landscape, documented software gaps, and a live, re-weightable matrix ranking ten project ideas across difficulty, usefulness, feasibility, fundability, and strategic fit.
Open the briefing Pitch deckThe Trusted Referee
An 11-slide pitch for the top-ranked idea — an open verification & benchmarking service for the AI-weather era, built on MET/METplus and the BEACON testbed. Press S for speaker notes.
Open the deckPresenter notes handout → — the deck's per-slide talking points, print-ready.
Working prototypes
Try the tools
Four self-contained, browser-native prototypes that make hazard, design, and computing science legible — no install, no server, runs offline.
Plume Sentinel
A tunable Gaussian-plume hazard viewer: a 2D danger-band plan view and a WebGL 3D volume, colored by concentration mapped to danger levels. Presets for an SO₂ stack, a chlorine leak, and wildfire smoke.
Launch the viewer Prototype · impactsEvacuation Decision Theater
A wind-driven wildfire (and hurricane) hazard coupled to a road-network evacuation model. Tune the winds, the rain, and when people leave — and watch the single road out gridlock, or everyone get out in time.
Launch the simulation Prototype · solar designShadow Study & Solar Design
Where the sun is, and where shadows fall, at any place, date, and time of year — for window placement, passive-solar gain, and orientation for fastest snow melt. Plan, isometric-3D, and sun-path views with a winter-sun exposure heatmap.
Open the tool Prototype · awarenessAI Compute Footprint
What does one AI response actually cost in energy, water, and carbon? Tune the assumptions and watch the range move — an honest "it depends," shown as low–central–high with a cited methodology, not a single scary number.
Open the estimatorDeep dives
The project write-ups
Per-idea working documents: what to build (scope, architecture, phasing, risks) and how to position & fund it. Markdown — download or open as text.
Research
An honest investigation
Can fast neural weather models be made to handle 100-year storms? An ML-expert agenda, a devil's-advocate teardown, and the reconciled verdict.
The finding, in one line: for genuinely unprecedented events the honest lever isn't retraining the emulator — it's a fast ML model that knows when it's out of its depth and hands off to physics, with the handoff verified. Which is exactly the trusted-referee thesis, applied.