Technology
Trading technology built for systematic US stocks
We develop internal systems across the full research-to-execution stack—market data, event intelligence, risk weighting, post-trade analysis, and production controls.
Our technology is purpose-built for systematic trading research—not adapted from generic portfolio management platforms. Components are modular, documented, and tested so engineers and researchers can work in parallel without fragile dependencies.
Machine learning is used selectively where it adds measurable value: supporting classification, weighting, anomaly detection, and post-trade analysis. Models are versioned, evaluated against held-out data, and subject to the same risk constraints as rule-based logic. We do not treat ML output as authoritative without validation and human oversight.
Market Data Infrastructure
Reliable, versioned market data is the foundation of every research workflow and production system we run.
We ingest tick, bar, and reference data for US stocks into pipelines designed for reproducibility. Each dataset is timestamped, validated, and stored with clear lineage so researchers can rerun experiments against the same inputs months later.
Normalisation layers handle corporate actions, symbol changes, and session boundaries. Quality checks flag gaps, stale feeds, and anomalous prints before data reaches downstream models or execution services.
Key capabilities
- Unified storage for research and production workloads
- Versioned datasets with reproducible transforms
- Automated validation and feed health monitoring
- Support for microstructure and aggregated timeframes
Catalyst and News Classification
Structured event intelligence helps us separate signal from noise across earnings, macro releases, and corporate announcements.
News and event feeds are parsed, deduplicated, and mapped to instruments and catalyst types. Classification rules and models assign each item a category—earnings, guidance, M&A, regulatory, macro—and a relevance score relative to our active universe.
Human-readable labels and machine-readable tags coexist in the same pipeline, so researchers can filter events quickly while systematic strategies consume structured inputs directly.
ML application
Machine learning supports text classification, entity resolution, and relevance scoring—particularly where rule-based systems alone cannot scale across headline volume or linguistic variation.
Key capabilities
- Entity linking to US stock symbols and sectors
- Catalyst taxonomy aligned to trading research needs
- Latency-aware ingestion for time-sensitive events
- Audit trails from raw headline to classified output
Adaptive Risk Weighting
Position and portfolio weights respond to measured risk—not static assumptions about how markets behave.
Risk weighting modules combine factor exposures, volatility estimates, and regime indicators to adjust how capital is allocated across signals and instruments. Weights are bounded by explicit constraints so adaptive logic cannot override hard risk limits.
The system is designed for transparency: every weight change can be traced to the inputs and parameters that produced it, supporting review before and after deployment.
ML application
ML models may inform volatility forecasting, regime detection, and dynamic weight suggestions—but final bounds are enforced by deterministic risk rules, not model output alone.
Key capabilities
- Factor exposure tracking across volatility, sector, and momentum
- Constraint-aware weight adjustment within defined bounds
- Regime-sensitive scaling without unconstrained leverage
- Logging and replay for post-hoc review
Post-Trade Pattern Discovery
Execution quality and strategy behaviour are analysed systematically after each session—not guessed from aggregate P&L.
Trade logs, fill data, and market context are joined into analysis datasets that support slippage attribution, timing review, and signal decay measurement. The goal is operational insight: where execution diverged from expectation, and whether research assumptions held in live conditions.
Pattern discovery workflows surface recurring structures—time-of-day effects, spread sensitivity, catalyst reaction profiles—without requiring manual inspection of every trade.
ML application
Unsupervised and supervised ML techniques can assist with clustering trade outcomes, detecting anomalies in fill behaviour, and identifying patterns that warrant further research—always as inputs to human review, not autonomous strategy changes.
Key capabilities
- Slippage and implementation shortfall measurement
- Join of fills with contemporaneous market state
- Session-level and rolling aggregate reporting
- Hypothesis generation for research follow-up
Execution and Risk Controls
Production trading runs through services built for observability, fault tolerance, and hard risk enforcement.
Order management logic translates approved signals into live orders with explicit checks at every stage: position limits, notional caps, restricted instruments, and session rules. Alerts fire when behaviour approaches thresholds—not only when limits are breached.
Execution services are instrumented for latency, rejection rates, and partial fills. Risk controls operate independently of strategy code so a logic error in research cannot bypass capital safeguards.
Key capabilities
- Pre-trade risk checks before order submission
- Real-time position and exposure monitoring
- Circuit breakers and manual override capability
- Structured logging for compliance and debugging
The descriptions on this page reflect our internal technology focus areas and are provided for general informational purposes. They do not describe live fund performance, guaranteed outcomes, or investment advice. System capabilities evolve as our research programme develops.
Building systems with us
We collaborate with developers, ML engineers, and technology partners who share an interest in production-grade trading infrastructure. If that describes your work, we would welcome a conversation.