Skip to content
Architecture

A trading firm, wearing the shape of a machine.

Beneath the dashboard is a layered system: cognition, execution, risk, and coverage — engineered to make one decision well, many thousands of times a day.

01Cognition

The algo's brain

The core is an ensemble of specialized models — a regime classifier, a directional bias model, a volatility model, and a meta-controller that weighs the others.

Every five seconds, a fresh feature vector is computed: price action across multiple horizons, microstructure features from order-flow, realized and implied volatility, calendar pressure from upcoming events, and a slow-moving macro state.

The output is not a binary 'buy' or 'sell.' It is a confidence number between zero and one, attached to a target size. Below the confidence threshold, the algo does the most expensive thing in trading — it does nothing.

02Execution

The execution layer

Once a decision crosses the confidence threshold, it is handed to the execution layer. The routing logic shops liquidity in real time and selects the venue where the trade can be filled with the lowest expected slippage.

From signal to fill is sub-millisecond on most venues. Latency is not vanity here — at machine speed, the difference between a 0.5ms and 1.5ms round-trip can be the difference between a winning trade and an event-driven loss.

03Risk

Risk management

Risk is not a layer that sits on top — it is built into the kernel. Per-trade stops, per-day stops, volatility-adjusted sizing, news-window flattening, exposure caps per asset class, and a kill-switch tier that doesn't require human approval.

When realized volatility spikes past a threshold, position sizes shrink before the next decision is even considered. If a daily drawdown limit is hit, the algo flattens and the desk reviews before the system trades again.

04Coverage

Multi-asset engine

One algo runs across gold, FX majors, indices, and the major cryptocurrencies — sized as one portfolio. Correlations are estimated live, and the engine refuses to stack exposure in correlated directions even when individual signals call for it.

The benefit to the investor is non-obvious but powerful: a single account quietly diversifies across asset classes that retail traders almost always over-concentrate in.

Decision flow

Five gates. One decision. Repeated every market tick.

The flow

One decision. Five gates.

Every trade the algo takes passes through these five stages, in order, in under a millisecond. Scroll to walk the path.

For the investor

What you don't have to do.

Watch the chart at 3am
Time the entry, time the exit
Argue with a stop loss
Build a thesis on the news
Reconcile spreadsheets
Trust a single asset class
Pay a manager who underperforms
Hope, basically