AI/ML · Robotics · Distributed systems

Backend engineer, now teaching robots to walk.

I'm Dominik. Below is a Unitree Go2 simulated in MuJoCo, walking on a policy I trained. It runs entirely in your browser. Steer it with the keyboard, or switch its brain off and watch it fall.

Downloads about 6 MB (MuJoCo WebAssembly, robot meshes, policy weights).

Unitree Go2 · MuJoCo WebAssembly · policy trained with PPO

Full screen How it works

to train on one GPU (200M steps): 5.5 min; to train on one GPU (200M steps)
pushes recovered in testing: 120 / 120; pushes recovered in testing
per policy step, in your browser: < 1 ms; per policy step, in your browser

About

I've spent about ten years building backend and distributed systems: Go microservices, GraphQL, Kafka, AWS and GCP.

Now I'm applying that to machine learning and robotics. That means training policies in simulation, being careful about evaluation, and getting models to run in real time.

More about me →

Projects

github.com/domgolonka

ufc-fight-vision

Predicts UFC fight winners from style features measured in fight video and compares the predictions with prediction-market prices.

In progress: odds collector built, video pipeline next

Python
httpx
pytest

nautilus-alpha

A model-agnostic alpha research harness for NautilusTrader, built around leakage-safe evaluation and honest trial counting.

Python
NautilusTrader
NumPy
pandas

latencybot-rust

A Rust port of a Polymarket latency-arbitrage bot: Binance L2 feed, calibrated edge detection, paper trading and risk halts.

Rust
Tokio
WebSockets

direx

An open-source marketing workspace where AI agents draft campaigns (email, social, newsletter, SEO, video clips) and nothing goes out until a person approves it.

Open source, coming soon

TypeScript
Bun
Hono
Postgres
React
Vercel AI SDK