AI engine.
A model-serving layer, an LLM gateway, and the operational practice that maintains both in production.
Nova — machine-learning service
Tabular models (LightGBM) for signal generation and conviction scoring. Trained on historical data, walk-forward validated, exposed as REST endpoints, and monitored in production.
AiEngine — LLM gateway
A central HTTP gateway fronting multiple LLM providers. Routes by caller allowlist, handles streaming, tracks usage. Provider substitution is possible without application changes.
Vision models
Open-source vision models applied to filings, charts, and structured documents where the cost-benefit is justified.
Drift and economic monitoring
Models in production are monitored for accuracy, calibration, and economic impact. Models whose economic lift disappears are flagged for review.
Sentinel.
An autonomous options engine built on the same order, pricing and risk infrastructure as MyAllies. Sentinel takes the option ideas produced by our AI engine, turns them into orders and manages every position until it is closed or expires — without manual intervention.
- ✓ From signal to order: each candidate is checked against live, time-stamped quotes before an order is placed.
- ✓ Commission-aware exits: apart from a manual close by the owner, a position is bought back only when that is profitable after all commissions; otherwise it is held to expiry, where it may expire worthless, be assigned or close at a loss.
- ✓ Continuous reconciliation: every order, fill and position is matched against the broker's records; discrepancies are corrected automatically where the records allow it and flagged for review otherwise.
- ✓ Fail-closed by design: stale data, missing quotes or a broker refusal skip a trade instead of guessing.
Status: running as a pilot on a broker paper-trading account. Sentinel is an in-house engine; it does not manage client money and is not offered as an investment service.