Centenum Labs is the research arm of Centenum Technologies. We work at the intersection of neurosymbolic computation, program synthesis, and causal modelling. Building AI that can be audited, corrected, and trusted at the level the decisions demand.
Combining learned representations with explicit, inspectable rules. Systems that can be audited, corrected, and trusted because their reasoning is inspectable by construction, not by post-hoc explanation.
Turning intent into executable logic rather than probabilistic completion. Machines that construct verifiable programs from specifications, examples, and structured intent.
Reasoning about what causes what, not just what correlates with what. Building systems that represent interventional structure so their predictions survive changes in the world they were trained on.
Learning mathematical expressions directly from data. Symbolic regression as a bridge between raw ML and interpretable, human-readable models, especially in scientific and high-stakes domains.
The dominant bet in AI today is that optimising for likelihood is enough. Train a system to predict the next token across enough data, at enough scale, and a system you can trust with consequential decisions will emerge.
We are building on the assumption that it will not.
Reliable, auditable, domain-specific AI will not be trained into existence. It will be engineered, at the intersection of neurosymbolic computation, program synthesis, and causal modelling. Our parent company, Centenum Technologies, has shipped production AI systems since 2018. That ground truth shapes everything we build in the lab.
This is the compressed version. The full argument is set out in our founding thesis, Likelihood Is Not Truth.