Technology

Under the hood of a
self-built trading platform.

One event-driven core powers research and production alike. This is how market data becomes signal, signal becomes orders, and orders become fills — deterministically, and at scale.

Event-drivenNanosecond clockSim = LiveMulti-asset

System Architecture

Four layers, one message bus.

Strategies sit on top. Beneath them, an engine core routes every message through a single deterministic event bus, drawing from the data engine and dispatching to venue adapters — simulated or live.

STRATEGY LAYER

Momentum

Stat-Arb

Market-Making

ML Alpha

ENGINE CORE

◆ EVENT BUS · deterministic message queue ◆

Clock

Portfolio

Risk Engine

Execution

DATA ENGINE

Feed handlers

Normalizer

Point-in-time store

VENUE ADAPTERS

SIM

Equities

Futures

Crypto

The Event-Driven Core

Everything is an event.

Quotes, trades, bars, orders and fills all flow through one queue, ordered by a monotonic nanosecond clock. Replay the same stream and you get the same result — every time.

QUOTEorder book & top-of-book updates

BARtime / tick / volume aggregates

ORDERsubmit / modify / cancel commands

FILLexecutions & position deltas

engine/loop.py

while self.clock.running:
    evt = self.bus.next()      # ns-ordered
    match evt.kind:
        case QUOTE | BAR:
            self.strategy.on_data(evt)
        case ORDER:
            self.risk.check(evt)
            self.venue.route(evt)
        case FILL:
            self.portfolio.apply(evt)

Data Pipeline

Point-in-time, or it didn't happen.

Every field is stamped with the moment it became knowable. Research reads the world exactly as it looked then — eliminating look-ahead bias by construction.

LATENCY BUDGET · feed → strategy

Ingest & decode~0.4 ms

Normalize & enrich~0.6 ms

Feature compute~0.8 ms

Dispatch to strategy~0.2 ms

Illustrative budget — figures are representative of the pipeline design, not a performance claim.

PB+

Historical tick, quote & book depth

Versioned

Reproducible datasets, immutable snapshots

Continuous

Automated quality & anomaly checks

Execution & Risk

Every order runs the same gauntlet.

Pre-trade risk, execution algorithm and post-trade attribution apply identically in simulation and production. The order lifecycle is a single, auditable state machine.

Intent

Strategy emits target

Risk check

Limits, exposure, kill-switch

Execution

Schedule & child orders

Fills

Venue confirmations

Attribution

P&L, cost, slippage

Owned end-to-end.
Built to be reproduced.

The same platform that explores an idea is the one that trades it. That is our edge, and our discipline.

Explore the AI Atlas ↗Get in touch