AI Trading

Automation with the guardrails on

Research, execution and risk tooling for desks that want models in production, not in a slide deck.

The premise

A model is only as good as the plumbing under it

Most automated strategies do not fail because the signal was wrong. They fail because costs were understated, limits lived in the model, or nobody noticed drift for six weeks.

Our suite is built around those three failure modes: honest cost modelling, server-side risk, and telemetry you can audit.

Illustration of a signal model connected to a market chart

Capabilities

Four tools, one mandate

Signal research

Hypothesis tooling over 12 years of tick data with survivorship-safe universes and point-in-time fundamentals.

Execution algorithms

Participation, TWAP and liquidity-seeking schedules tuned per venue from our own fill history.

Risk automation

Volatility-aware position ceilings and automatic de-risking the moment a limit is breached.

Model monitoring

Live drift detection against back-test expectations, with alerting into your own channels.

The stack

From notebook to matching engine

Zurich colocation available

  1. 01

    Research

    Python / notebook sandbox, tick + L2 archive

    batch

  2. 02

    Signal service

    Versioned model artefacts, feature store

    < 5 ms

  3. 03

    Risk gateway

    Server-side limits, kill switch, pre-trade checks

    < 180 µs

  4. 04

    Execution

    FIX 4.4 / WebSocket, venue-aware routing

    < 900 µs

  5. 05

    Telemetry

    Fill attribution, drift metrics, audit log

    streaming

180 µs
Pre-trade risk check
12 yrs
Tick and L2 archive
99.98%
Gateway uptime
24/5
Model support cover

How it works

Four stages from mandate to scaled capital

  1. 01

    Define the mandate

    Universe, holding period, risk budget and hard constraints, agreed in writing.

  2. 02

    Back-test honestly

    Costs, slippage and borrow modelled from our own fill data, not idealised prices.

  3. 03

    Shadow deploy

    The model trades in parallel with no capital for a full cycle and is scored against expectation.

  4. 04

    Scale deliberately

    Capital increases in steps, each gated on realised versus expected performance.

Transparency pledge

Four commitments, written into the mandate

Request a mandate discussion
  • No model is presented with a back-test that excludes transaction costs.
  • Every automated order is attributable to a named model version.
  • Risk limits are enforced server-side, never inside the model.
  • You keep your intellectual property; we never trade against your signals.