Areas of work

Signal models

Tabular models (LightGBM, XGBoost) for predictive signals across equities, options, and macro factors. Walk-forward validated.

Conviction and ranking

Cross-sectional and time-series conviction scores, operated in production.

Anomaly detection

Detection of pricing errors, missing data, corporate-action issues, and suspicious flow.

LLM workflows

Earnings-call summarisation, filing extraction, news classification, entity resolution. Processing throughput exceeds 40,000 documents per day.

Vision and document AI

Document extraction, image-based filings, chart interpretation using open-source vision models adapted to financial workflows.

Feature engineering

Domain-specific features — earnings surprise, insider clustering, supply-chain shocks, options skew — engineered for financial use cases.

Operating philosophy

Every model in production is instrumented for drift, error calibration, and economic impact. Kill switches are mandatory before client-facing deployment.

Where a problem lacks adequate data, a clean target, or sufficient economic signal, we will say so.

Open-source and frontier models are selected per workload according to cost, latency, and accuracy requirements.

ML engagement enquiry.

Describe the model or workflow under consideration. We will respond with feasibility and scope.

Contact