From research we build the tools that make AI more reliable.

Industrial Probabilistic Calibration

We develop advanced metrics (Brier, ECE, MCE, NLL, Sharpness, Resolution) and recalibration methods to guarantee reliable decisions in critical contexts.

An AI is only as reliable as its calibration.

Metacognition and Uncertainty Honesty

We study how an agent can know when it does not know, recognise its limits, declare uncertainty and avoid harmful overconfidence.

We work with measures of the self-knowledge gap, semantic perturbations and structural verification.

Evolutionary Optimization and Digital Entities

We apply evolutionary algorithms to surface emergent behaviour in digital entities and organisms, in order to improve prompts, tool chains, reasoning sequences and decision policies.

We look for systems that evolve towards greater coherence.

Multi-LLM Governance and Cross-Validation

We research how multiple models can correct one another: weighting structures, metacognitive divergence, verified ensembles and hybrid decision paths.

Regulated Critical Domains

We create datasets and protocols specific to sectors where error is intolerable: education, justice, finance, government, mining, heavy industry.

Geometries of Hidden States

We go below the observable answer. We analyse hidden states, trajectories, perturbations and latent geometry. What structures can we measure when a model holds an identity, receives perturbations and must regulate its own trajectory?