- One research stack across aligned intelligence, neural interfaces and post-silicon compute
- Pre-registered experiments, reported whether or not they succeed
- A public claims register: each claim with its evidence and what would falsify it
From synapse to silicon. NeuroGenesys is a frontier technology research lab decoding the human brain, engineering aligned general intelligence, and studying the post-silicon substrate both will run on. It is based at SHACK15, atop San Francisco's Ferry Building.
NeuroGenesys is a research lab, not a product company. Its output today is published research. It runs no interface trials, and the Genesys SDK is a concept preview with an early-access waitlist. Nothing below describes a shipping model, implant or chip.
The Problem: Capability Is Outrunning Understanding
Intelligence research, neuroscience and chip design are usually pursued apart, and each runs into limits the others could help explain.
- Frontier models get more capable faster than anyone can verify what they are doing
- Neural interfaces are gaining electrodes quickly, and decoding, not hardware, is becoming the bottleneck
- AI compute is approaching a power wall, and claims for alternative hardware are rarely tested against honest baselines
The Approach: One Research Stack
NeuroGenesys treats these as one continuous loop: decode the brain, model what is learned, and build the compute it runs on, keeping the method strict enough that capability never outruns understanding.
Research Programs
Genesys. The path to continual-learning world models that reason with test-time compute and improve without forgetting, grounded in published results on verifier exploitability and catastrophic forgetting.
Cortex. How foundation-model decoders generalize across sessions, subjects and tasks, with mental privacy as a first-class constraint.
Computational neuroscience. Connectomics and brain foundation models.
SynapseOne and Lumen. Neuromorphic and photonic programs built on published cost-model campaigns that map where analog in-memory compute and all-optical inference hold up, and where both collapse.
How the Work Is Done
Experiments are pre-registered, run once and reported whether or not they succeed, against baselines at equal budget. A public claims register lists each surviving claim with its evidence and what would falsify it.
What Happens Next
The papers are published in the research archive at neurogenesys.si, which is also where to request early access to the Genesys program.
