Field notes
AI SystemsQuantitative Engineering

Designing AI-Assisted Quantitative Systems

Where AI belongs—and where it does not—in strategy design, validation, and execution.

Separate assistance from authority

An assistant may translate natural language into a candidate strategy, explain indicators, or suggest exploration ranges. Every result should resolve into a deterministic configuration that users can inspect, test, and approve.

Make simulation reproducible

Backtests need versioned inputs, stable assumptions, explicit fees, and visible failure states. Intelligence without reproducibility creates persuasive noise rather than engineering evidence.

Design for refusal

Strong systems know when not to proceed. Invalid configurations, incomplete market data, risk-limit conflicts, and model uncertainty must become explicit product states—not hidden logs.

Draft article placeholder: expand with reviewed examples and evidence before promoting publicly.