NeuroGenesysThe science of minds.
The engineering of intelligence.
NeuroGenesys is one continuous research stack — from frontier world models and cortical interfaces down to the neuromorphic, photonic, and custom silicon they will ultimately run on.
Four flagship programs.
One research loop.
We study the brain, model what we learn, and chart the silicon those principles could run on. Explore the programs, from field notes and pre-registered experiments to cost models.
GenesysFrontierModels
Our research path to continual-learning world models. It builds on two ideas: when each skill sits in its own frozen expert, forgetting reduces to choosing the right expert, and an exploitable verifier limits how far extra samples help. Pre-registered tests confirmed the first on Split-CIFAR-100 and found no Best-of-N decline with real judges by N = 32. The program page sets out each claim, its evidence level and where it fails.
Explore ProgramCortexNeuralInterfaces
The field has reached 4,096-electrode arrays and conversational-rate speech decoding. Our Cortex program follows the decoder layer in our field notes — how foundation models generalize across sessions, subjects, and arrays.
Explore ProgramSynapseOneNeuromorphic
Before committing to silicon, we modelled where it wins: analog in-memory compute holds an 11.8× decode advantage for state-space models, and a hybrid keeps 5.6–6.8× only with sliding-window attention. SynapseOne starts from that published rule.
Explore ProgramLumenPhotonics&ASICs
Our falsification study mapped the honest envelope of all-optical inference — near parity against an ASIC, but with 3000× precision robustness and zero-forgetting frozen features. Lumen studies where light actually wins.
Explore ProgramA lab platform that
runs its own experiments.
A concept preview of the Genesys SDK — the research copilot the program is designing: safety guardrails injected in real time, decoders tuned as they drift, every experiment inside its sandbox.
Real-time Safety Injection
Run agentic workloads without fear. As you type, the platform intercepts over-permissive or unsafe configurations, injecting strict guardrails and sandbox policies before anything reaches a live model.
import { Agent, Model, Oversight } from '@neurogenesys/sdk';
import { Logger } from '@neurogenesys/logger';
const logger = new Logger({ level: 'debug' });
logger.info('Initializing Genesys agent cluster...');
// ── [01] INITIALIZE AI AGENT ──
const analyzer = new Agent({
name: 'NeuralData_Analyst',
model: Model.Genesys1,
oversight: Oversight.SCALABLE,
memory: true
});
Autonomous Workspace
Forget manual configuration. The platform analyzes your experiment continuously, writing calibrated decoder settings and environment variables directly into your workspace.
Lab Copilot
Talk to your experiments. Genesys Copilot acts as a research engineer on call — watching decoder drift, compute budgets, and safety envelopes, and resolving issues before they cost you a session.
