About AutoSim
Generating a model was never the hard part.
AutoSim exists because the bottleneck in serious simulation moved. Writing the code stopped being expensive; proving the code deserves to be believed never did.
How it started
Built from the work of defending models in front of people paid to break them.
AlphaIQ was built around one kind of engagement: agent-based models of markets, portfolios and institutions, delivered to teams whose decisions get challenged — by a risk committee, a regulator, a board, or a counterparty. In every one of those rooms the same thing happened. The model was rarely the argument. The argument was whether the model could be trusted.
That work produced a repeatable answer: specify the system precisely, fit it to data with a method that reports its own uncertainty, test it on data it never saw, and write down what it cannot tell you. Done by hand, that takes a specialist team weeks per model. Done well, it survives scrutiny.
Then general-purpose AI made the first half nearly free. Anyone can now ask a model for a simulator and get something that reads convincingly. Almost nobody gets a way to check whether it is right. That is the failure mode AutoSim was built for: the confident, unvalidated model that looks finished.
So AutoSim automates the whole method, not the easy part of it. The AI reads your description and writes the simulator. Executed code — real Python running in your browser — fits the parameters, scores the model out of sample, and produces the artifacts you would need to defend it. Eighteen months of that validation engineering is the product.
What the product refuses to do
Five commitments the pipeline holds.
01
Executed code decides, never narrative
AI proposes structure and priors. Every number that reaches you comes out of a computation you can re-run, not out of a sentence.
02
No score is folded together
Fit, validation and identifiability are reported separately. A model that fits well and generalises badly is shown as exactly that, not averaged into a single grade.
03
Honest failure over confident output
If the system cannot be identified from the data you brought, AutoSim says so and narrows the claim it is willing to make.
04
Every stage leaves an artifact
Specification, priors with citations, posteriors, out-of-sample results and known limitations are all readable, exportable and dated.
05
Your compute, your data
Simulations run in your browser through WebAssembly. There is no upload step for the numbers themselves and no queue to wait in.
Who it is for
Anyone who has to defend a number.
- Quant and risk teams who already build models and want the validation layer to stop being a manual project.
- Analysts and strategists with the domain question but no simulation team behind them.
- Consultants and researchers who need a model, the evidence for it, and a document that survives review.
- Institutions under model-risk oversight who need the audit trail as much as the answer.
The method is published
The book is the method. AutoSim is the lab.
AutoSim ships alongside Agentic Simulations: A Practitioner's Handbook and its companion volume — seventeen worked models, all runnable inside the product with cited priors. If you want to see the reasoning behind a pipeline stage before you trust it, it is written down.
Ilan Gleiser, founder of AlphaIQ
AlphaIQ
AutoSim is how AlphaIQ ships its method.
AlphaIQ builds calibrated, validated agent-based simulations of the markets, portfolios and institutions our clients operate in. AutoSim is the self-serve product; bespoke engagements are the hands-on path for teams who want the model built with them.
Next step
Bring a system you need to defend.
Describe it in plain language and AutoSim will build, calibrate and score a simulation of it while you watch. If the question needs more than the product alone, AlphaIQ will build it with you.