Aiman Shabib/ Bonn, Germany

N 50.7° · E 7.1°

2026 · v1.0aiman.tech

[ R-001 / IDX ]Researcher of intelligent systems

Engineerof intelligentsystems

Trying to understand the physics of intelligence. Currently poking at Parallax - an attempt at a different substrate for cognition, somewhere between attractor dynamics and central workspace theory.

Attractor dynamics · CWS
Continual & adaptive cognition
Geometry of representations

Open Parallax

Read the notes

aiman@shabib.net→

Scroll/R-001 · parallax

01Manifesto

I love to code - not in pursuit of efficiency, but as an art of creation

Trying to understand the physics of intelligence. Currently working on Parallax - an attempt at a different substrate for cognition, somewhere between attractor dynamics and central workspace theory - and keeping a notebook of smaller side-tracks along the way.

Years on the engine

Unity · VR · MR · shipped

Main research line

Parallax

CWS · attractor · energy

Side projects

dendrite · pi · neuroevo · ucp

[ Stack · live trace ]18 entries · always evolving

PyTorch·Transformer architectures·Inference & deployment·Continual learning·Central workspace theory·World models·Geometry of representations·Custom inductive biases·Evolutionary algorithms·TypeScript / Node·Rust·C# / .NET·Python·Unity Engine·VR / Mixed Reality·Compute & VFX shaders·NLP · TTS · STT·Agentic systems·PyTorch·Transformer architectures·Inference & deployment·Continual learning·Central workspace theory·World models·Geometry of representations·Custom inductive biases·Evolutionary algorithms·TypeScript / Node·Rust·C# / .NET·Python·Unity Engine·VR / Mixed Reality·Compute & VFX shaders·NLP · TTS · STT·Agentic systems·

/Agentic systems/NLP · TTS · STT/Compute & VFX shaders/VR / Mixed Reality/Unity Engine/Python/C# / .NET/Rust/TypeScript / Node/Evolutionary algorithms/Custom inductive biases/Geometry of representations/World models/Central workspace theory/Continual learning/Inference & deployment/Transformer architectures/PyTorch/Agentic systems/NLP · TTS · STT/Compute & VFX shaders/VR / Mixed Reality/Unity Engine/Python/C# / .NET/Rust/TypeScript / Node/Evolutionary algorithms/Custom inductive biases/Geometry of representations/World models/Central workspace theory/Continual learning/Inference & deployment/Transformer architectures/PyTorch

02Domains of practice

Three threads, one substrate. They feed each other - real-time discipline informs the ML, the ML reshapes the product, the product demands new tooling.

[01]

Open ML research

I read modern ML as physics - representation geometry, inductive biases, the structure of the bottleneck. I'm building toward continual, adaptive cognition rather than ever-bigger static models.

Parallax · central workspace, continual learning, world models
Dendritic Unit · activation-free MLP block
Notes on grokking, attention, emergence

[02]

AI in production

Where the research meets the customer. B2B platforms running autonomous AI in places people pay for it - voice campaigns, channel-agnostic chat, eval loops, the boring infrastructure that holds it up.

Monamy · autonomous CRM running phone + chat campaigns
Channel-agnostic agent runtimes & guardrails
Production inference, deployment, observability

[03]

Real-time & immersive

Years inside Unity teach you to respect the frame budget. I bring that discipline back to the AI side - and forward into agentic tooling for the engines themselves.

Shipped VR titles on Meta Quest 2 & 3
Mixed-reality prototypes · XR Hack winner
UCP · agent-native control plane for the Unity Editor

03Selected work

8 entries · 2021 - 2026

[R-001]

Research2024 - present

Parallax

A modality-agnostic cognitive core that settles into a metastable regime in a strange-attractor landscape. Cognition decoupled from token-clocked autoregressive generation. Reasoning depth at O(1) memory.

PyTorchCUDACustom kernelsCWSDEQ-flavoured

Open case↗

[R-002]

Research2026

Dendritic Unit

A side track. What if you took the activation function off the MLP and let the unit itself be nonlinear, through learned competition between dendrite-style branches.

PyTorchCUDATransformers

Open case↗

[T-003]

Tooling2026

Unity Control Protocol

A Rust CLI + C# editor bridge that exposes Unity to agents over JSON-RPC. For the AI-native Unity workflow.

RustC#WebSocketJSON-RPC 2.0

Open case↗

[P-004]

Product2024 - present

Monamy

An AI-native CRM running fully autonomous mass campaigns and channel-agnostic support - phone, chat, anywhere. monamy.de.

TypeScriptPythonNLPTTS · STTAgentic runtime

Open case↗

[S-005]

Shipped2024

Orbit Brawl

Mixed-reality space brawler that turns your living room into a battleground. Built in 48 hours. 1st place gaming, XR Hack Cologne.

UnityMixed RealityMeta Quest

Open case↗

[R-006]

Research2025

Neural Evolution

Self-driving cars in Unity, evolved without supervision. Custom neural net + population-based training, configurable from the inspector. Generalises to unseen tracks under pure evolutionary pressure.

UnityC#Custom NNEvolutionary algorithms

Open case↗

[S-007]

Shipped2021 - 2023

Everslaught Invasion

Fast-paced VR action title shipped on Meta Quest 2 & 3. Years on the engine side reworking and optimizing core systems.

LIYA

Redesigned an AI-driven onboarding & interview flow - unbiased candidate profiling, auto-generated company-side dossiers. myliya.com.

LLMsConversational AIEval loops

Open case↗

See the full portfolio

04Open lab

Architectures · inductive biases · in-flight

Most architecture choices are inherited accidents. The lab is where I question the ones that became defaults

L-001Active · primary research line · v3

Parallax

A Central Workspace as a continuous-state dynamical system that settles into a metastable regime in an attractor landscape, perturbed by a heterogeneous pool of expert probes. Reasoning is settling. K loop is decoupled from the token loop. Candidate replacement for the residual stack itself.

CWSAttractor dynamicsDEQ-flavouredEnergyContinual

Open page↗

L-002Active · own track

Dendritic Unit

A research aside. Take the activation function out of the MLP block and let the unit itself be nonlinear, the way biological dendrites are - through learned competition and coactivation between branches.

MLPStructural plasticityCompetition

Open page↗

L-003Internal · in-flight

Pi

A focused experiment exploring what kind of inductive bias could replace attention for a narrow class of long-context, structure-rich tasks.

AttentionLong contextArchitecture

Internal · ask

L-004Public · maintained

Neural Evolution

Self-driving agents in Unity, evolved from scratch. Custom NN, population-based training, configurable from the inspector. A playground for reading the geometry of behavior.

EvolutionBehaviorReal-time

Read on GitHub↗

Open the lab

05Notes & essays

Long-form thinking · low-frequency

Research dispatches and theory notes from the lab. Most of them are attempts to make my own thinking visible to me. If they end up making it visible to someone else, even better.

Published15 min

Note 001 · Architecture

Parallax: cognition without the token clock

What I have been working on for almost two years across three rewrites. A small recurrent core that thinks by settling, language treated as one of several sensory channels, and reasoning depth detached from the token loop. Notes from the v3 codebase and what I am still trying to figure out.

ParallaxCWSAttractor dynamicsDEQArchitecture

Read →

Published8 min

Note 002 · Parallax · K-DOF

The model stops and ponders

One observation from training the v3 CWS that I keep showing people. The model autonomously escalates compute on hard tokens, without ever being told what 'hard' means. Five to six times more reasoning iterations on harder lookups, paid for at constant memory.

ParallaxCWSK-DOFTest-time compute

Read →

Published13 min

Note 003 · Honest log

Parallax: what I got wrong three times

Parallax v3 is the third honest restart. v1 was a RAG replacement that hit a ceiling no projector architecture could lift. v2 was a dual-process system with a memory bank that worked best when it collapsed to a single vector. v3 is what I have now. Notes from getting the project wrong and learning more from each version than from any one of them being right.

ParallaxProcessHonest log

Read →

Published8 min

Note 004 · Surprises

Memory collapse is a feature

Halfway through training a memory-augmented architecture, I watched the entire memory bank collapse to a single vector. I spent a month trying to fix it. Then I looked at the numbers, realised the collapse was actually the optimum, and unlearned a thing I thought I knew about memory.

Parallaxv2SurprisesInductive bias

Read →

All notes

06Contact