Featured work
BrainCore getbraincore.com
in progress (testing)
A human brain in software, built region by region from published neuroscience as a biologically detailed spiking neural network (SNN). No LLM, no pretrained weights, no backpropagation.
- Go
- Spiking neural network (SNN)
- Leaky integrate-and-fire
- Potjans–Diesmann microcircuit
- STDP
- Computational neuroscience
- Neuromorphic-ready
The goal
BrainCore (getbraincore.com) is building a complete human brain in software, region by region, from published neuroscience. Finished, it should see, hear, remember and learn from its own experience. It is a brain simulation, not a language model: no LLM, no pretrained weights, no backpropagation.
How it is built
- Neurons — leaky integrate-and-fire, integrated with the Rotter–Diesmann exact scheme.
- Cortex — columns follow the Potjans–Diesmann cortical microcircuit.
- Learning — spike-timing-dependent plasticity (STDP), dopamine and sleep mechanisms are built.
- Engine — written in Go, bit-exact deterministic: a SHA-256 hash over every spike, so a run is either identical or it is a bug.
- Speed — 89.7M synaptic events per second on a laptop CPU. The signal is spikes, which suits neuromorphic chips.
What works today
- Vision — a model retina feeds area 17. It reads 4 letters at 94 %; the preregistered acceptance rule passed on 20 of 20 seeds.
- Hearing — a Meddis cochlea feeds area 41. Tonotopy holds on 20 of 20 seeds.
- Next — hippocampus is in progress.
Status
Research stage. Paper 1 was frozen on 11 September 2026: Retinotopic vision and tonotopic hearing. Raising pre-seed.
Before the brain
The name started on a different project: github.com/vbcherepanov/braincore, local-first cognitive memory for AI coding agents in pure Go (MCP-native, attention, decision graph, the right to say “I don’t know”). Its spiking-network memory experiment, SubStat, was folded into BrainCore. Memory for coding agents lives on in total-agent-memory.