Finding structure in places that resist it.

Aleatoric Lab was created out of a fascination with finding structure in places that resist it, and finding honesty in systems that are usually rewarded for projecting confidence they do not have. What drives the work here is a particular kind of curiosity: the belief that understanding something properly means being willing to say exactly how much you do not know, and that the most useful models are not the ones that give you an answer but the ones that tell you how much to trust it.

The name comes from the Latin alea, meaning the roll of a die, and it refers specifically to aleatoric uncertainty, which is the irreducible randomness that lives inside every complex system regardless of how much data you collect or how sophisticated your methods become. What makes this idea compelling is that it does not belong to any single domain. It lives equally inside the decisions a hospital makes under resource pressure, inside a supply chain absorbing a geopolitical shock, inside a financial model being stress-tested against scenarios that have never happened before, and inside social systems trying to serve populations whose needs cannot be fully anticipated. The mathematics of irreducible uncertainty transfers across all of them, and what changes is only the domain knowledge you bring to bear on it.

At Aleatoric Lab we build probabilistic ML and Bayesian systems that take this seriously, developing methods and tools for the domains where being honest about uncertainty is not a methodological nicety but an operational necessity. Because a model that knows its own limits is not a weaker model, it is a more trustworthy one, and in high-stakes environments that distinction is everything.

The mathematics of uncertainty does not respect disciplines.

Machine learning is at its most powerful not when it memorises patterns but when it generalises, when the same underlying structure surfaces across problems that look nothing like each other on the surface. The strength of probabilistic methods in particular lies in exactly this kind of transfer: the mathematics does not care whether you are modelling a hospital, a supply chain, or a social network.

A Gaussian process modelling flood risk and one modelling misinformation credibility are solving the same mathematical problem. Multi-agent systems for supply chains and multi-agent systems for adversarial reasoning share the same core challenge. What connects the research areas at Aleatoric Lab is not the domain but the problem structure.

Current focus areas include Bayesian inference and Gaussian processes for prediction under uncertainty, causal and epistemic reasoning for robustness and stress-testing, and neurosymbolic methods for combining logical structure with probabilistic learning. Application domains span supply chain, financial systems, health and social policy, misinformation, disaster risk, and multi-agent systems.

What we are working on, now.

Currently live. New work follows.

For research collaborations, consulting enquiries, or anything else, please get in touch directly.