AeroTwin
Join the waitlist →
AEROTWIN · OPS SCOPE
PROJ EQUIRECTANGULAR
78°N — 58°S
180°W ⟷ 180°E
SEED 0x5EED1A7C   TICK 000000   FLIGHTS 0   ON-TIME   IRROPS 0
LIVE · DETERMINISTIC

Airline Operations · Digital Twin

Every slot, delay, and diversion — modeled to the tick.

AeroTwin is a deterministic digital twin of an airline's operations center. Schedules, weather, irregular operations, cost — direct the whole operation, reproducible to the cent.

What it is

A simulation only worth running if it's honest.

The closer the twin gets to a real operations center, the more its decisions — and their costs — actually mean something.

The Surface — an operational sandbox

Run the operation.

A serious airline-operations simulation, not a tycoon game. Schedules and slots, aircraft and cost, weather that turns. Irregular operations arrive without warning, and you make the cancel, delay, and recovery calls a real ops controller lives with.

The Depth — a real model

Modeled to the tick.

Underneath: A-CDM flight timing, weather from a seeded stream, integer-cent money. The same mechanics a real ops center runs on — reproducible to the cent, replayable from the command log.

Domain fidelity

It speaks the operation's language.

Not a weather-and-money abstraction wearing airline paint. The model is built around the artifacts an ops center actually works from — and it keeps moving toward the ones it doesn't have yet.

Modeled today
  • A-CDM flight timing — the milestone clock an ops center runs on
  • A seeded weather stream that drives delays and disruption
  • Slots, schedule, and cancellations under pressure
  • Integer-cent cost attached to every decision
On the roadmap
  • IATA SSIM schedule ingestion
  • Adverse-weather scenarios from historical METAR / TAF
  • SID / STAR and airspace constraints
  • BADA-style aircraft performance
  • Fuel price, and LCC-vs-FSC P&L

How it holds up

Built on a backbone that doesn't drift.

CORE / 01

Deterministic core

Same seed, same command log, same world — down to the last cent. Every run is replayable and auditable, not a one-off you can never reproduce.

replay → final-world hash matched
OPS / 02

Irregular operations

Weather, delays, and cancellations arrive from a seeded stream — disruptive but reproducible. They cascade through the schedule and the cost line exactly as they would in a real operation.

weather → seeded sub-RNG
OPEN / 03

Open core

The simulation engine is open source. Inspect the model, run it headless with nothing but cargo test, and write plugins against a semver-stable ABI.

crates/sim_core · plugin_abi

Why deterministic matters

Replayable isn't a technicality. It's the whole point.

"Same seed, same world" reads like an engineering footnote. In an operation it means the counterfactual is exactly reproducible — the one thing a real ops center never gets.

For the enthusiast

The snowstorm that paralyzed the hub — the same weather, the same schedule, waiting for you. Try your recovery, watch it cascade, rewind, and take a different call. As many times as it takes.

For the professional

Post-mortem the cancel-versus-divert decision against identical conditions every run, and read the cost of each path to the cent. A sandbox where the what-if is reproducible, not an anecdote.

The backbone, in spec

Determinism, spelled out.

Time
sim sees only
tick: u64
No wall-clock in the logic. Pace is a UI concern.
Money
i64 cents
Integer accounting. No float rounding drift.
Randomness
rng ← (seed,
producer, tick)
A weather delay is reproducible, not luck.
Verification
content-hash
replay
The final world is hash-checked on every build.

Models in the loop

Frontier models, wired into the operation.

Not a chatbot bolted to the side. Language models drive the parts of an operation that are messy, human, and never the same twice — so the pressure on the desk feels real.

SIGNAL / 01

Snags & comms

Pilot-reported snags and virtual-ATC exchanges, written in the voice of the moment — the ambiguous, incomplete inputs a controller actually has to act on.

SIGNAL / 02

Passenger fallout

When a bank collapses, models read passenger sentiment and weigh rebooking cost against the clock — so an IRROPS call carries its human and financial weight.

SIGNAL / 03

Your choice of model

Bring the frontier model you trust — plug it into the loop and see how it holds up under an operation that doesn't wait.

GPT · Claude · Gemini · Grok

Early access

Be in the tower when it opens.

We're building in public. Join the waitlist for first access, the dev log, and the opening set of operation templates.

NO SPAM · DEV LOG + LAUNCH ONLY · UNSUBSCRIBE ANYTIME