GPU-native evolutionary AI framework

Real-time evolutionary intelligence for Unreal Engine.

NeuroFlux Genesis is a real-time evolutionary intelligence framework that combines neurodynamics, perception, and GPU-native execution from sensing to action.

Built for investor diligence and developer studios that need transparent, end-to-end AI infrastructure inside UE5.

GPU-first runtimeBio-inspired learningUE5-native plugin

Built by Coheron Technologies Kft

What NFG is

A full-stack framework for real-time evolution.

NFG replaces static training with continuous adaptation inside the simulation loop.

What sets it apart

  • Evolution and learning run continuously in real time.
  • Neurodynamics (STDP + neuromodulation) shape behavior online.
  • GPU-native data flow from perception to action.
  • Unreal Engine 5 plugin with tooling and telemetry.

Breakthroughs

Breakthroughs that define NFG

Selected milestones that are genuinely differentiating.

GPU-native evolution stack

CMA-ES eigendecomposition, crossover and mutation kernels, and Hall of Fittest orchestration for large populations.

MMS neurons and TS-ReLU

Per-neuron state machines with evolvable dynamics to express adaptive behavior beyond static layers.

Neurodynamics in real time

STDP plasticity combined with four-channel neuromodulation (DA/NE/ACh/5-HT) to shape learning.

Perception built for agents

GPU raycasting plus retina-style feature extraction for autonomous perception.

World prediction and what-if rollouts

Multi-step GPU rollouts with action variants to score outcomes in real time.

NFGServer distributed learning backbone

Node.js + MongoDB backend that persists models and telemetry, validates uploads, and orchestrates training queues for cross-world learning.

Architecture

End-to-end GPU pipeline

Data stays on the GPU from sensors to actions.

1

Sensors and perception

Raycasting and vision features generated directly on the GPU.

2

Neurodynamics

MMS neurons, STDP, and neuromodulators shape activity in real time.

3

Evolution and selection

CMA-ES, crossover, mutation, and Hall of Fittest orchestration.

4

Action and physics

Actions feed directly into the UE5 simulation loop.

5

Telemetry and control

NFGServer persists models and telemetry and orchestrates training queues across worlds.

End-to-end GPU pipeline (simplified).

Capabilities

Capabilities in clusters

Transparent view of what is implemented today.

Evolution engine

Optimization and selection at scale.

  • GPU CMA-ES with eigendecomposition
  • GPU crossover and mutation kernels
  • Hall of Fittest persistence and insertion
  • Adaptive mutation scheduling strategies

Neurodynamics

Online learning inspired by biology.

  • STDP plasticity for causal learning
  • Four neuromodulator channels (DA/NE/ACh/5-HT)
  • Predictive coding and error-driven signals
  • Sleep replay consolidation

Neural architecture

Expressive, evolvable neuron design.

  • Multi-Mode State (MMS) neurons
  • TS-ReLU with evolvable parameters
  • GPU LSTM/GRU/Elman support
  • Per-layer activation selection

Perception and world modeling

Agent-first sensing and prediction.

  • RT-core raycasting sensors
  • Retina-style GPU vision pipeline
  • World predictor rollouts
  • Proprioception and physics sensors

Runtime and tooling

UE5-native compute with deep control.

  • RDG compute pipeline
  • Async compute scheduling via RDG
  • Config and serialization system
  • Telemetry, debug, and visualization tools

NFGServer & distributed learning

Persistence, telemetry, and orchestration beyond a single world.

  • REST/WebSocket model uploads with validation
  • MongoDB persistence for models, runs, and telemetry
  • Training queues for distributed workers
  • Shared evolution database across environments

Full feature catalog

Explore the complete public catalog of implemented NFG capabilities.

View catalog

Use cases

Built for studios and investors

NFG targets demanding real-time environments and large agent populations.

No public benchmarks yet. Full feature catalog available publicly.

Adaptive game AI

NPCs and agents that evolve against player strategies in real time.

Simulation and robotics R&D

Train control policies in high-fidelity UE5 scenes with rich sensor input.

Multi-agent experimentation

Large populations for emergent behavior and strategy discovery.

Technical demonstrator

AIRacing

A physics-heavy lab designed to stress NFG perception and evolution.

Private builds and technical deep dives available on request.

View demo

Company

Coheron Technologies Kft

Independent R&D focused on real-time evolutionary AI.

NFG is built inside the AIRacing lab environment and engineered for production-grade UE5 integration.

Sole creator

Architecture, GPU backend, and UE5 integration are designed and implemented by the founder.

Feature catalog with implementation references available for diligence.

Contact

Investor and studio inquiries

Reach out for technical diligence, private builds, and roadmap discussion.

Email is the fastest path to a deep dive.

Email Coheron
contact@coheron.ai