cleoanka

an interactive lab

cleoanka

I build simulations, learning systems and native tools for Apple Silicon.

↓ EXPLORE scroll down

about

I simulate to understand

I think the best way to understand something is to simulate it from scratch: orbits, combustion chambers, quantum circuits, learning neurons. Then I try to turn what comes out into a tool people can actually use.

I report results by measuring them. A negative result is still a result; my docs say what didn't work as plainly as what did. Click any project title below and a small interactive page opens where you can poke at the idea yourself.

learning
Backprop-free and continual learning; networks that don't forget, with local rules.
physics
Physics engines and numerical methods: N-body, combustion, quantum.
vision
Computer vision and road safety: detection, tracking, speed, plates.
apple silicon
Native software in Metal, Swift and Rust.

work

Things you can poke at

Every figure is running live. Click a title to open the project's interactive page.

1

Simulation & physics

3 projects
FIG. 1 — Three bodies on the figure-eight orbit (Chenciner–Montgomery).

PYTHON

aura-volterra-engine

An N-body engine that solves the whole trajectory at once instead of stepping with RK4: damped Picard iteration of the Volterra integral form, finite-speed gravity and optional 1PN precession.

FIG. 2 — An 8-state quantum walk. Frequent measurement freezes it: the Zeno effect.

RUST

zeno

An Apple Silicon-native quantum circuit simulator, compiler and runner: NEON acceleration, noise channels, OpenQASM 3. Slice time finely enough and the arrow never moves.

FIG. 3 — Pressure–volume loop of a four-stroke engine cycle.

C++

redline

A native-Metal build of engine-sim, the real-time combustion engine simulator, for Apple Silicon macOS, with live engine swapping and gamepad support.

2

Learning systems

3 projects
FIG. 4 — Spike raster: background activity and replaying cell assemblies.

PYTHON

aura-mind

A brain-like spiking network that learns online and grows itself. No backprop, only STDP and three-factor learning, and every step is measured against forgetting.

FIG. 5 — Two labels diffusing over a k-NN graph.

PYTHON

atlas

Few-label learning on a manifold: with a good metric one label per class recovers about 95% of MNIST, with a poor one it hurts. A visual study of edge purity and phase transitions.

FIG. 6 — Order book depth and mid price.

PYTHON

kairos

A dual-process trading brain: strictly causal order-book perception (System 1) fused with multi-agent LLM reasoning (System 2). It closes the backtest look-ahead hole by construction.

3

Vision & road safety

2 projects
FIG. 7 — Tracking, speed and rule violations (limit 90 km/h).

PYTHON

panoptes

A road and vehicle intelligence platform: multi-stream detection (YOLO26, RF-DETR), clean-room ByteTrack, calibrated speed, Turkish plate reading, a declarative rules engine, FastAPI and Docker/K8s.

FIG. 8 — The camera catches an event and requests priority bandwidth over 5G.

PYTHON

RoadGuard

5G and AI road safety for TEKNOFEST 2026: vehicle, plate, driver-behaviour and speed detection with YOLO26, talking to the 5G network through the CAMARA QoD/NV APIs.

4

macOS tools

3 projects
FIG. 9 — Note graph and semantic search.

RUST

aura-app

An AI-native, local-first Markdown second brain for macOS. Your notes never leave the device.

FIG. 10 — An 8-band equalizer over a live spectrum.

SWIFT

auricle

Per-app volume, EQ, boost and output routing: a menu bar sound mixer built on Core Audio process taps.

FIG. 11 — Search, find, launch.

SWIFT

cosmodrome

A native remake of the full-screen app grid macOS Tahoe removed: a fast, keyboard-driven Launchpad.

5

Interactive writing

1 project
FIG. 12 — Scratched bits get repaired by an error-correcting code.

JAVASCRIPT

unscratchable

Scratch this message. It heals. An interactive essay on error-correcting codes, from parity bits to Reed–Solomon, with zero dependencies.

by the numbers

A year, counted

14

Public repositories on GitHub.

by primary language

    start dates

    contact

    Got an idea? Write

    A joint project, an internship, a question or just hello: these are the fastest ways to reach me.