Overview Research Performance Technology About The market doesn't We built a system that doesn't either. Autonomous AI research, alternative data, and institutional-grade risk management — running around the clock. Explore Results → How It Works Multi-Layer AI Eight distinct model classes working in concert — from sequence learning to reinforcement learning — dynamically reweighted as market conditions shift. Alternative Data Beyond price. We process regulatory filings, patent activity, earnings linguistics, and macro flows into quantifiable signals before they appear in prices. Statistical Rigour Every signal is independently validated before entering the system. We apply the same statistical standards as institutional research desks. Out-of-sample.
Walk-forward validated. All metrics computed on unseen data from 2023–2024. Training window: 2018–2022. Results reflect realistic slippage and commission assumptions. CPCV N=6 Deflated Sharpe 1,000-path Monte Carlo Walk-Forward 8 Windows Return Profile Detailed Metrics Market Structure Sharpe Ratio vs SPY 0.76 Sortino Ratio downside-adjusted Ann. Return OOS 2023–2024 Max Drawdown trough to peak Calmar Ratio return / max DD Profit Factor gross P&L ratio Cumulative Return vs Benchmark Monthly Returns 24 months Risk-Adjusted Sharpe Ratio (OOS) Sortino Ratio Calmar Ratio Max Drawdown Monte Carlo 95% CI Ann. Return 95% CI Trade Statistics Total Trades (OOS) Win Rate Profit Factor Best Asset META +45.7% Avg Trade Duration 3.2 days Validation Method Ensemble Weights deep learning gradient boost Capital Assets Rebalance Kill Switch Market Regime Distribution Drawdown Profile Trend Up Trend Down Volatile Mean Revert Statistical Confidence 1,000-path
Monte Carlo
bootstrapped. 95% confidence interval Sharpe Ratio Annualised Return Max Drawdown Now running on Alpaca. Live trading started May 2025. $100K starting capital across 10 liquid US equities. Live results published as they accumulate. Fifteen phases.
One coherent system. Each layer adds compounding intelligence. What cannot be replicated is the calibration, the integration, and the signal combinations that emerge over time. A system designed for institutional rigour from the ground up. Architecture built on peer-reviewed research. Every decision references academic literature. Every component is independently testable and replaceable. The system grows more intelligent over time — a weekly discovery loop finds, validates, and integrates new signals without human intervention. Axon Intelligence was built on the principle that genuine alpha is earned through disciplined research — not intuition. The system emerged from a conviction that institutional-grade quantitative research should be rigorous and statistically validated — regardless of the scale at which it operates. Every backtest result is walk-forward validated. Every signal is independently tested before entering the system. The architecture is self-improving by design.

Überblick Forschung Performance Technologie Über uns Der Markt wartet Wir haben ein System gebaut, das es auch nicht tut. Autonome KI-Forschung, alternative Daten und institutionelles Risikomanagement — rund um die Uhr. Ergebnisse ansehen → Wie es funktioniert Mehrschichtige KI Acht verschiedene Modellklassen arbeiten zusammen — von sequentiellem Lernen bis zu Reinforcement Learning — dynamisch gewichtet nach Marktbedingungen. Alternative Daten Über Preise hinaus. Wir verarbeiten regulatorische Meldungen, Patentaktivitäten, Sprachanalysen und Makro-Flows zu messbaren Signalen. Statistische Strenge Jedes Signal wird unabhängig validiert, bevor es ins System aufgenommen wird. Dieselben statistischen Standards wie institutionelle Research-Desks. Out-of-sample.
Walk-forward validiert. Alle Kennzahlen auf ungesehenen Daten 2023–2024 berechnet. Trainingszeitraum: 2018–2022. Ergebnisse berücksichtigen realistische Slippage- und Kommissionsannahmen. CPCV N=6 Deflated Sharpe 1.000-Pfad Monte Carlo Walk-Forward 8 Fenster Renditeprofil Detaillierte Kennzahlen Marktstruktur Sharpe Ratio vs SPY 0,76 Sortino Ratio Abwärtsrisikobereinigt Ann. Rendite OOS 2023–2024 Max. Drawdown Tiefpunkt zu Hochpunkt Calmar Ratio Rendite / Max DD Profit-Faktor Brutto-G&V-Verhältnis Kumulative Rendite vs. Benchmark Monatliche Renditen 24 Monate Risikobereinigt Sharpe Ratio (OOS) Sortino Ratio Calmar Ratio Max. Drawdown Monte Carlo 95% KI Ann. Rendite 95% KI Handelsstatistik Gesamtanzahl Trades (OOS) Trefferquote Profit-Faktor Bestes Asset META +45,7% Ø Handelsdauer 3,2 Tage Validierungsmethode Ensemble-Gewichtung Deep Learning Gradient Boosting Kapital Assets Rebalancing Kill-Switch Marktregime-Verteilung Drawdown-Profil Aufwärtstrend Abwärtstrend Volatil Mean Reversion Statistische Konfidenz 1.000 Pfade
Monte Carlo
bootstrapped. 95% Konfidenzintervall Sharpe Ratio Annualisierte Rendite Max. Drawdown Läuft jetzt auf Alpaca. Live Trading gestartet Mai 2025. $100K Startkapital in 10 liquiden US-Aktien. Live-Ergebnisse werden veröffentlicht sobald sie vorliegen. Fünfzehn Phasen.
Ein kohärentes System. Jede Ebene fügt zusammengesetzte Intelligenz hinzu. Was nicht repliziert werden kann: Kalibrierung, Integration und die Signalkombinationen die sich über Zeit entwickeln. Ein System, das von Grund auf für institutionelle Strenge gebaut wurde. Architektur auf Basis von peer-reviewter Forschung. Jede Entscheidung referenziert akademische Literatur. Jede Komponente ist unabhängig testbar und austauschbar. Das System wird mit der Zeit intelligenter — eine wöchentliche Entdeckungsschleife findet, validiert und integriert neue Signale ohne menschliches Eingreifen. Axon Intelligence wurde auf dem Prinzip gebaut, dass echter Alpha durch disziplinierte Forschung verdient wird — nicht durch Intuition. Das System entstand aus der Überzeugung, dass institutionelle Quantitative-Research-Standards rigoros und statistisch validiert sein sollten — unabhängig vom Maßstab. Jedes Backtest-Ergebnis ist Walk-Forward validiert. Jedes Signal wird unabhängig geprüft. Die Architektur ist von Natur aus selbstverbessernd.

The market doesn't

We built a system that doesn't either. Autonomous AI research, alternative data, and institutional-grade risk management — running around the clock.

1.484

Sharpe Ratio

15.75%

Ann. Return

2.342

Sortino

Equity Curve — OOS 2023–2024

+15.75% Annualised · Sharpe 1.484

Ann. Return 15.75% · Monte Carlo Validated · Sharpe 1.484 · Sortino 2.342 · 15 Research Phases · 102 Modules · 8 AI Models · Walk-forward Validated · Calmar 1.687 · Ann. Return 15.75% · Monte Carlo Validated

Multi-Layer AI

Eight distinct model classes working in concert — from sequence learning to reinforcement learning — dynamically reweighted as market conditions shift.

Technology →

Alternative Data

Beyond price. We process regulatory filings, patent activity, earnings linguistics, and macro flows into quantifiable signals before they appear in prices.

Research →

Statistical Rigour

Every signal is independently validated before entering the system. We apply the same statistical standards as institutional research desks.

Performance →

Performance

Out-of-sample.
Walk-forward validated.

CPCV N=6 Deflated Sharpe 1,000-path Monte Carlo Walk-Forward 8 Windows

All metrics computed on unseen data from 2023–2024. Training window: 2018–2022. Results reflect realistic slippage and commission assumptions.

1.484

Sharpe Ratio

vs SPY 0.76

2.342

Sortino Ratio

downside-adjusted

+15.75%

Ann. Return

OOS 2023–2024

9.33%

Max Drawdown

trough to peak

1.687

Calmar Ratio

return / max DD

1.195

Profit Factor

gross P&L ratio

Return Profile

Cumulative Return vs Benchmark OOS 2023–2024

Monthly Returns 24 months

Detailed Metrics

Risk-Adjusted

Sharpe Ratio (OOS)1.484

Sortino Ratio2.342

Calmar Ratio1.687

Max Drawdown9.33%

Monte Carlo 95% CI[0.48 – 2.48]

Ann. Return 95% CI[7.6% – 22.5%]

Trade Statistics

Total Trades (OOS)4,023

Win Rate26.1%

Profit Factor1.195

Best AssetMETA +45.7%

Avg Trade Duration3.2 days

Validation MethodCPCV · DSR

Ensemble Weights

Transformer deep learning

29%

LightGBM gradient boost

28.3%

LSTM deep learning

27.7%

XGBoost gradient boost

15%

Live

Now running on Alpaca.

Live trading started May 2025. $100K starting capital across 10 liquid US equities. Live results published as they accumulate.

$100K

Capital

Assets

Daily

Rebalance

<3%

Kill Switch

Market Structure

Market Regime Distribution OOS 2023–2024

Trend Up66.8%

Trend Down22.2%

Volatile7.3%

Mean Revert3.2%

Drawdown Profile OOS

Statistical Confidence

1,000-path
Monte Carlo
bootstrapped.

Sharpe Ratio

0.4895% confidence interval2.48

Annualised Return

7.6%95% confidence interval22.5%

Max Drawdown

5.1%95% confidence interval18.4%

Research

Fifteen phases.
One coherent system.

Each layer adds compounding intelligence. What cannot be replicated is the calibration, the integration, and the signal combinations that emerge over time.

Ph. 01–02 Data & Feature Engineering

data_manager

Cached market data pipeline for 20-asset universe. Efficient storage with automatic freshness checking.

feature_engine

68+ technical and microstructure features. Captures market memory, volatility regimes, and hidden structure in price data.

Ph. 03–04 Model Suite & Signal Generation

ensemble

Four core models combined via rolling Sharpe attribution. Weights shift automatically as models gain or lose predictive power.

regime_detector

Four-state market classification. Signal blend adjusts per regime — momentum strategies run in trending markets, mean-reversion in ranging ones.

meta_labeling

A secondary filter that learns when to trust the primary signal and when to stay out. Structurally reduces overfitting.

dynamic_weighting

Signals with negative recent Sharpe are automatically disabled. The system self-curates its own signal set over time.

Ph. 05–08 Risk & Execution

tail_risk

Stress-tested against five historical crisis scenarios. Tail losses modelled using extreme value theory — not normal distribution assumptions.

correlation_risk

Time-varying correlation model. Automatically reduces leverage when cross-asset correlations spike — a hallmark of crisis periods.

smart_routing

Orders scheduled to minimise market impact. Trade size calibrated to daily volume so the system avoids moving prices against itself.

compliance

Seven pre-trade checks run before every order. Immutable audit trail with complete lineage from signal to fill.

Ph. 09–11 Advanced Alpha & Alternative Data

microstructure

Estimates the probability that institutional, informed money is actively trading in an asset. High readings trigger position reduction.

insider_signal

Parses live SEC regulatory filings. Cluster buying by multiple insiders within a short window generates a strong directional signal.

earnings_nlp

Earnings call transcripts scored for linguistic confidence and uncertainty. Management tone often leads price action by days.

tda_features

Applies mathematical topology to price sequences. Detects structural shifts in market behaviour before they become visible in conventional indicators.

Ph. 13–15 Self-Improving Intelligence

gene_pool

Discovers new alpha formulae by evolving mathematical expressions over market data. No human specifies what to look for.

auto_researcher

Weekly loop: discover → validate → check for overlap with existing signals → integrate if it genuinely adds new information.

shadow_mode

New system versions run silently alongside production. Statistical test gates live promotion. Automatic rollback on underperformance.

nas_search

Model architecture is evolved, not designed. Each run produces a different neural network — the result cannot be reproduced without running the search.

Technology

A system designed for institutional rigour from the ground up.

Architecture built on peer-reviewed research. Every decision references academic literature. Every component is independently testable and replaceable.

The system grows more intelligent over time — a weekly discovery loop finds, validates, and integrates new signals without human intervention.

Phase 01–04

Intelligence Layer

Eight model classes — temporal, graph-based, probabilistic, and reinforcement learning — combined in a dynamically weighted ensemble that adapts in real time.

Deep LearningGraph Neural NetsRL Agent

Phase 05–08

Risk Engine

Time-varying correlation modelling, extreme value tail risk, factor exposure decomposition, and optimal trade scheduling to minimise market impact.

DCC-GARCHTail RiskSmart Routing

Phase 13–15

Self-Improvement

Genetic programming discovers new signals. Neural architecture search evolves model structure. Weekly validation loop gates everything before it goes live.

Genetic ProgrammingNASAuto-Research

Machine Learning

PyTorchNeural networks

LightGBMGradient boosting

XGBoostTree ensemble

HuggingFaceNLP models

Quantitative

cvxpyOptimisation

scipyStatistical fitting

statsmodelsGARCH · HMM

OptunaHyperparameter search

Infrastructure

Alpaca APILive execution

APSchedulerJob scheduling

SQLAlchemyTrade storage

ParquetData caching

Validation

CPCVPurged cross-val.

Deflated SharpeMulti-test adj.

Monte Carlo1,000-path

Walk-forward8 windows

About

Rigour over speculation.
Always.

Axon Intelligence was built on the principle that genuine alpha is earned through disciplined research — not intuition.

The system emerged from a conviction that institutional-grade quantitative research should be rigorous and statistically validated — regardless of the scale at which it operates.

Every backtest result is walk-forward validated. Every signal is independently tested before entering the system. The architecture is self-improving by design.

Established2025

FocusQuantitative Research & Algorithmic Trading Systems

Architecture15 phases · 102 modules · 8 AI model classes

UniverseUS equities — large-cap and ETFs

StatusLive via Alpaca

ValidationCPCV · Deflated Sharpe · 1,000-path Monte Carlo

DisclaimerResearch only. Not financial advice.

Out-of-sample.Walk-forward validated.

Fifteen phases.One coherent system.

A system designed for institutional rigour from the ground up.

Rigour over speculation.Always.

Out-of-sample.
Walk-forward validated.

Fifteen phases.
One coherent system.

Rigour over speculation.
Always.