Research & AI Work

Artificial Adaptive Intelligence proposes a distinct stage between today's narrow AI systems and hypothetical general intelligence.

Rather than defining progress purely through model scale, the work focuses on adaptivity: the ability of a system to reduce its dependence on human-specified parameters, architectures and configuration while maintaining competitive performance across different tasks.

The monograph connects ideas from meta-learning, AutoML, continual learning, evolutionary computation and physics-informed modelling into a broader framework for measuring how much intelligence a system can internalise rather than requiring humans to specify.

Multi-agent AI systems often waste significant resources by repeatedly transmitting raw text between agents and recomputing information that has already been processed.

Q-KVComm explores a different approach: allowing agents to communicate through compressed internal representations.

The framework combines adaptive KV-cache quantisation, information-preserving compression and cross-model calibration, achieving approximately 5-6x communication compression while maintaining semantic fidelity in the reported experiments.

ELENA explores how evolutionary algorithms can become more adaptive rather than relying on fixed optimisation behaviour.

Inspired by biological epigenetics, the framework introduces an adaptive memory system that dynamically influences mutation, crossover and stability during optimisation.

The research was evaluated across network optimisation problems, with the strongest improvements emerging on larger and more complex problem spaces.

Traditional gradient-boosting systems use tree structures and splitting behaviour that remain essentially fixed throughout training.

MorphBoost introduces self-organising decision trees that dynamically modify their splitting behaviour as learning progresses.

The framework combines adaptive split criteria, automated problem fingerprinting and changing optimisation pressure to allow the learning architecture itself to respond to the problem being solved.

ORCA investigates how machine learning can identify changing system regimes rather than treating relationships between variables as static.

Applied to financial markets, the framework combines spectral graph theory, random matrix theory and supervised learning to analyse how correlation structures evolve over time.

The broader research question is how an intelligent model can recognise that the environment itself has changed and adjust its interpretation accordingly.

AlphaJet demonstrates end-to-end intelligent optimisation in a highly constrained engineering environment.

Starting from a textual mission specification, the system generates and evolves aircraft concepts while evaluating aerodynamics, structure, stability, packaging and geometric constraints.

Instead of using AI simply to suggest an answer, AlphaJet closes an iterative loop traditionally managed by human experts: generate, evaluate, adapt and improve.