Researching
My current research investigates how intelligent systems can preserve knowledge, coordinate complex processes, form interpretable representations, and evolve over time without losing coherence.
DREAM: Dynamic and Persistent Memory
DREAM explores persistent and adaptive memory for intelligent agents. The research treats memory not as static storage, but as a dynamic system involving relevance, consolidation, influence, revision, forgetting, and long-term behavioral effects.
The current version investigates dynamic equilibrium, semantic drift, cognitive immunity, meta-memory, controlled forgetting, memory governance, and the distinction between preserving information and allowing it to influence future decisions.
My present work focuses on strengthening DREAM’s theoretical foundations, defining measurable behaviors, and designing controlled experiments that expose both the architecture’s capabilities and its limitations.
MQ-AGI: Asynchronous Cognitive Coordination
MQ-AGI investigates the use of asynchronous queues to coordinate cognitive processes in intelligent systems. Instead of requiring every operation to occur in a single sequential flow, the architecture separates tasks into specialized processing queues.
The research examines how perception, memory, reasoning, evaluation, and action can operate with different priorities and execution times while remaining part of a coherent system. Its broader goal is to explore modularity, concurrency, fault isolation, and cognitive coordination in more complex AI architectures.
LFD: Principles for Evolving Systems
LFD is a principle-oriented framework for designing systems that must change without becoming structurally opaque or internally inconsistent.
The research organizes architectural concerns into a set of principles addressing evolution, boundaries, responsibility, traceability, and long-term comprehensibility. It investigates how explicit design principles can support both conventional software systems and experimental intelligent architectures as their behavior and internal complexity grow.
HGC-Net: Hybrid Conceptual Representation
HGC-Net investigates hybrid neural architectures that combine explicit concepts with latent representations. Its purpose is to explore whether a model can preserve the flexibility of learned features while exposing meaningful conceptual structures that humans can inspect and intervene in.
Current experiments compare the architecture with strict concept bottleneck models and evaluate classification performance, concept quality, intervention behavior, representation separation, and visual explanations. The research also examines whether the model truly learns useful concepts or merely produces labels that appear interpretable.
Building
I am developing prototypes, benchmarks, and evaluation tools to test these ideas under controlled conditions. This work includes reference implementations for my research, experimental interfaces, visualization tools, and infrastructure for comparing architectures and documenting results.
I am also expanding Devmora UI, a reusable interface system for software projects, and maintaining CodeByMP as the public platform where I document research, engineering projects, technical experiments, and the reasoning behind them.
Other active engineering projects include Raygon, an experiment in contract-based system design, and GrimleyTK, a collection of tools exploring desktop applications, browser technology, and systems development.
Studying
I am currently studying machine learning, data science, cognitive architectures, memory systems, concept-based learning, experimental methodology, and software architecture.
A central part of this process is learning how to design stronger evaluations: defining falsifiable hypotheses, selecting meaningful baselines, isolating variables, measuring long-term behavior, and documenting negative results alongside successful ones.
My goal is not only to build architectures that appear promising, but to understand why they work, where they fail, and under which conditions their underlying assumptions remain valid.