The framework behind every engagement.
Our own AI models help develop your investment edge into a systematic strategy. The client receives a full license to the Eigen code used as part of the project.
Eigen: the direction a system cannot rotate.
Applied to most inputs, a system changes both direction and size. A few inputs come back pointing the same way, changed only in size. That multiple is the eigenvalue, and it is the clearest read available on what a system actually does.
- A
- The system, applied
- v
- The direction that survives it
- λ
- The scale it produces
A = [3 1; 1 3] · v = (1, 1) · λ = 4
The two directions A cannot rotate.
DIRECTION PRESERVED, MAGNITUDE SCALED
The framework is named for that standard. We look for the structure in a strategy that survives being applied to live capital, then measure exactly how much of it you keep.
Signal layer
Directional signals per security and per horizon. We use our own AI models for signal detection and integrate your rule sets, data, and insights into a systematic strategy.
Risk envelope
Distributional forecasts showing downside, median, and upside ranges. Position sizing and risk budgeting based on downside, not just direction.
Regime control
A meta-model that estimates how reliable each signal is in current conditions. Less exposure when conditions historically weaken the model.
Portfolio construction
Position sizing, entry and exit averaging, allocation strategies, time-horizon overlays. An implementable portfolio, not a list of predictions.
Risk management
Stops, risk parameters, correlation and systemic-risk monitoring built into the framework.
Execution and position management
Orders in broker formats, automated or manual execution, order netting, reconciliation to broker reports.
Audit and compliance
Time-stamping, immutable storage, SEC-compliant records. Evidence auditors and regulators recognize.
Explainability
Model-specific explain tools. Every decision can be traced to its drivers.
Why standard AI fails in financial markets, and what we do differently.
Most systematic approaches treat financial data as a generic time series and try to find patterns that repeat. The problem is that markets are not driven by patterns. They are driven by how investors interpret events, and the same event means something completely different depending on the macro environment it occurs in.
Rising unemployment in a low-rate environment is typically negative, meaning lower economic activity. In a high-rate environment the same data point can be positive, because it opens the path to rate cuts. A model that cannot account for that context will fail when conditions shift, regardless of how well it performed in backtesting.
Eigen is built around this reality. Our own AI models learn each instrument's sensitivity to rates, credit, volatility, and macro conditions. A classifier estimates direction and conviction. A quantile model estimates the plausible return range. A regime overlay assesses how reliable the signal is right now. We apply that capability to each client's investment logic, not as a standalone product.
And a training discipline that does not cheat: walk-forward validation, no look-ahead, no data leakage, quarterly retraining on a fixed rolling window with frozen models between cycles. What you see in testing is what has a realistic chance of working live.