Neuromodulatory Control Networks (NCNs), a novel LLM architectural modification inspired by the neuromodulatory systems in the vertebrate brain.

axon

Biologically inspired language model using Jaccard Surprise as its only training signal. No backprop. No GPU. Online Hebbian learning from corrections. Two-layer cortex with apical feedback. Runs on CPU under 200MB.

Modular Forward-Forward Network with independent processing modules and central coordinator. CIFAR-10: 68.65%.

A Python package implementing Rectified Spectral Units (ReSUs), a biologically inspired neural building block for backprop-free training using spectral decomposition and Canonical Correlation Analysis (CCA).

A real-time simulation of a virtual creature controlled by a spiking neural network built on biologically accurate mechanisms. The creature navigates a 2D world, forages for food, avoids predators and poisons, and learns from experience through dopamine-modulated synaptic plasticity.

A biologically-inspired AI with 2.34B virtual neurons, real-time vision, hearing, memory, and emotion — running fully local on GPU

BioGPT: A biologically-inspired dendritic language model with cortical columns, NMDA temporal memory, and pruning resilience that beats Pythia-160M

Bio-inspired manycore thermal management: decentralized O(1) per-node load routing (92.1% thermal variance reduction) and an ENAQT-analog circuit topology proposal. Reproducibility code for paper "Decentralized Thermal-State Load Routing..." (Costa, 2026). DOI: 10.5281/zenodo.19857070