Research notes
Notes on financial machine learning, quantitative methods, and building intelligent systems that people can trust.
Jul 22, 2026
Apex Arena is complete: a production Formula racing experience with resilient data handling, replayable race rooms, and a verified delivery path.
Jul 8, 2026
How to operate an adapted model with evaluation gates, versioning, monitoring, and a disciplined rollback path.
Jun 17, 2026
A decision guide for selecting an adaptation strategy based on task shift, cost, inference constraints, and operational needs.
Jun 3, 2026
A separate project: a Python routing agent that uses deterministic solvers first, then gated local inference, then a larger-model fallback.
May 20, 2026
Before choosing LoRA or full tuning, define the task, data lineage, quality thresholds, and non-goals.
May 6, 2026
How Apex Arena turns live Formula racing data into a bounded, evidence-linked conversation between five specialist agents.
Apr 15, 2026
A test strategy for agentic systems that measures tool use, abstention, recovery, and reviewer experience—not only final answers.
Mar 25, 2026
Why financial agents should be built as controlled workflows with segregated duties and explicit approval gates.
Mar 4, 2026
A safety-first view of agentic workflows for clinical support, where tools, approvals, and audit trails define usefulness.
Feb 18, 2026
How leakage, costs, and regime change turn a promising financial backtest into a fragile production strategy.
Jan 28, 2026
A practical framework for pairing predictive credit models with explanation, fairness review, and human judgment.
Jan 7, 2026
Why financial ML must be designed around decisions, uncertainty, and controls—not a single accuracy score.