/about

This blog is about my work on Sabohat — a brain-inspired AI model. The current architecture is a Spiking Neural Network (SNN); unlike classic ANNs, neurons here communicate not via continuous signals but through spikes, like in a biological brain.

The project started as a Graph Neural Network (GNN) — a 4-level (character → morpheme → word → semantic) graph structure. After several experiments I switched to SNN, since it is closer to the biological brain and more energy-efficient. I'll write about that transition in separate posts.

Main goals: (1) bring the model to a state where it can learn on its own, (2) significantly reduce catastrophic forgetting in continual learning.

So far 200+ experiments have been run; each tests spike encoding, plasticity rules and memory consolidation strategies.

Here I share: dev journals, technical notes, and open thoughts on language and AI.

// contact: hello@solvie.me