Exploring physical continuous-time neural circuits, biological connectomes, and ultra-low-latency neuromorphic computing to build safe, deterministic autonomous systems.
A 500-neuron Drosophila connectome executed via 30 Euler numerical integration steps in CPU L1 cache for sub-15 µs involuntary safety locks and least-privilege tool masking.
Columnar phase-shift ring attractors derived from FlyWire CX volumes to prevent cognitive drift across deep multi-file codebase refactoring.
Spike-Timing-Dependent Plasticity in a 10,000-dimensional sparse coding space to retain developer idioms without RAG or token stuffing.
Sparse matrix emulation of 139,255 biological neurons and 54.5 million synapses mapped at nanometer electron microscopy resolution.
A local 100 Hz connectome executive that handles AST navigation and safety on-device, querying cloud LLMs solely for brief creative code blocks.
Empirical evaluation showing AST-bound symbol isolation reduces context token overhead by 74% and eliminates out-of-scope edits across 1,000 repositories.
