ETM Logo
ETM-Network / The Neural Project
System Operational

The Path to AGI.

Starting with everyday apps. Scaling to Superintelligence.
We build state-of-the-art models for the devices you use today, to fund the intelligence of tomorrow.

> init_neural_engine()

Loading modules... [OK]

Optimizing tensors for edge... [OK]

> current_focus

["NeuralAudio", "Neural-TTS", "Neural-STT", "AGI Research"]

> _

NeuralAudio

Crystal clear audio. Powered by Edge AI.

Experience state-of-the-art denoising running locally on your device. Remove background noise instantly without your data ever leaving your phone.

On-Device Inference Real-time Denoising Privacy First
NeuralAudio App Interface

NeuralFit

AI-Powered Discipline.

Our upcoming fitness revolution. NeuralFit uses next-gen pose estimation to track your form and lock distracting apps until you've completed your workout.

Computer Vision Pose Estimation Digital Wellbeing
NeuralFit App Interface

Neural-TTS

Human-like speech. Instant generation.

Transform text into natural-sounding speech with a state-of-the-art text-to-speech model. Choose from four distinct voices and enjoy near real-time generation—all processed locally on your device.

On-Device Inference 4 Natural Voices English Language
Neural-TTS App Interface

Neural-STT

State-of-the-art transcription. Completely private.

Convert speech to text with cutting-edge accuracy using a Conformer architecture model. Record live audio or import files—all transcription happens locally on your device with zero data leaving your phone.

On-Device Inference File Import Real-time Recording
Neural-STT App Interface

NeuralVoice

TTS-powered audiobook reader.

Experience your favourite books with natural, human-like narration. NeuralVoice uses a state-of-the-art offline diffusion model to generate high-quality audio from your EPUBs and PDFs, keeping your listening experience completely private.

On-Device Inference Diffusion Model Offline Audiobook
NeuralVoice App Interface
Research Preview

Adaptive Chimera System

Universal goals. Emergent learning. Dynamic spawning.

What is Chimera?

Chimera is our initiative to create adaptive super-neurons—compute units that operate on event streams with rich internal state, capable of local learning without full backpropagation while remaining compatible with PyTorch tensors for rapid iteration.

Chimera agents blend four core signals—Expected Reward, Information Gain, Pattern Value, and Collective Predictions—to pursue a universal objective in any environment. The system continuously discovers reusable patterns, validates them against new experiences, and dynamically spawns specialist units when fresh perspectives increase momentum.

Core Architecture

01

Event Packets

Standard containers carrying timestamps, payload vectors, optional metadata, and reinforcement signals—enabling asynchronous processing across the runtime.

02

Chimera Units

Maintain internal memory (short-term vector + longer-term latent factors), process incoming events via gating functions, and apply local learning updates using event payloads and feedback signals.

03

Runtime Manager

Routes events to chimeras, collects outputs, schedules learning passes, persists state, and provides hooks for visualization and instrumentation.

04

Structural Plasticity

Mechanisms for spawning new chimeras, pruning underperforming ones, and reassigning specializations based on observed data patterns.

Mathematical Foundation

Activation a_t = gate(event, memory) × transform(event)
Confidence σ(energy(event, memory))
Learning Hebbian outer product + decay
Threshold Moving average → target sparsity

Research Results

Chimera vs Baseline accuracy
Chimera training progress

Early experiments show Chimera consistently outperforming baseline approaches, with dynamic unit spawning adapting to task complexity in real time.