journal
research publications, technical essays, and studies exploring the frontier between art, science, and intelligent systems.
on the architecture of persistent agentic frameworks
exploring the design principles behind systems that maintain coherent identity across sessions — how memory, policy selection, and skill-based reasoning create agents that evolve rather than reset.
embedding models and the geometry of meaning
a study of how vector representations encode semantic relationships, and why the spatial metaphor of meaning has practical implications for retrieval-augmented generation.
the RUNES protocol — skill-based agent orchestration
introducing a framework for composable agent behaviors through declarative skill graphs, policy routing, and emergent task decomposition.
browser automation as a cognitive extension
how autonomous browser agents can serve as externalized reasoning — navigating, extracting, and synthesizing information in ways that augment rather than replace human judgment.
edge inference on constrained hardware
deploying quantized language models on Jetson Orin and Raspberry Pi — benchmarks, optimization strategies, and the surprising viability of local AI at the edge.
multi-model evaluation frameworks
designing systematic approaches to comparing LLM outputs across dimensions of accuracy, coherence, and creative reasoning — beyond simple benchmarks.
memory architectures for conversational agents
a comparative analysis of episodic, semantic, and procedural memory systems in long-running AI agents — what to remember, how to forget, and when to synthesize.