Two decades of selected work
Multi-strategy portfolio management & risk
Ran a $1.6 billion multi-strategy book, roughly $600M multi-quant, $700M factor, and $300M derivatives and volatility, from signal research through sizing, hedging, and risk control under an investment-committee mandate. Over two years the market-neutral sleeve’s information ratio rose from about 0.8 to 1.1 with gross exposure held flat. The gain came from pruning decaying signals, not from taking more risk.
Regime-shift models for factor exposure
Built macro-regime models that read the prevailing environment and conditioned factor exposures to it. Risk scaled up or down by regime rather than running every signal flat-out. That kept the book’s tilts aligned with what the market was actually rewarding at the time.
Derivative overlays for tail hedging
Designed index, volatility, and cross-asset hedging overlays that protected institutional portfolios against tail events without bleeding carry in calm markets. Hedges scaled by regime rather than staying permanently on, which kept the cost of protection close to the actual risk.
Alternative-data signal research
Turned unstructured sources such as filings, earnings transcripts, and news into structured, point-in-time factors for systematic strategies. The hard part was avoiding look-ahead and survivorship bias. A signal that looks good in research is worthless if it cannot survive that test live.
Factor construction research
Researched and built the factors themselves across value, momentum, quality, carry, and low-volatility, including how each one is defined, cleaned, orthogonalised, and tested before it earns capital. Every candidate was scored for decay, capacity, and crowding, then checked against realistic costs in regimes it was never fitted to.
Portfolio optimisation & capital allocation
Allocated capital and risk budget across strategies with risk-parity, mean-variance, and Kelly-style sizing rather than treating every signal as equally deserving. The job was deciding how much of the book each edge should earn, sized to its conviction and its volatility.
Global macro strategy
Took directional and relative-value positions across rates, FX, and commodities, conditioned on the prevailing macro regime. The edge sat less in calling the level and more in reading when the regime was about to turn.
Volatility & options relative value
Built volatility-surface, skew, variance, and dispersion trades that valued and hedged optionality instead of simply owning direction. Pricing the convexity correctly was what made a position safe to hold when the move finally came.
Stress testing & scenario analysis
Ran historical and Monte Carlo scenarios, VaR and CVaR, and severe liquidity-stress assumptions so positions were sized against the bad case, not just the central one. A frozen-funding, widening-spread world was treated as a scenario to budget for well before it arrived.
Research-to-decision automation
Led the team that built an agentic pipeline taking an investment idea from raw research to a sized, risk-checked, source-linked recommendation. It compressed a roughly two-week cycle to about two days. A person still approves or rejects each call; the system does the assembly, not the judgment.
AI-driven opportunity screening
Built the system that reads thousands of filings and transcripts to surface and pre-vet opportunities against explicit quantitative criteria. It raised the names one analyst can actively monitor by roughly tenfold. Every figure links back to its source sentence. A second model checks each extraction before it is shown.
Self-directed AI research agents
Built research agents that run their own literature and data sweeps, critique their own output, and surface only what survives that check for human review. They take on the mechanical breadth of research so the team’s time goes to judgment.
RAG over financial filings
Built retrieval pipelines that answer questions over 10-Ks, transcripts, and loan indentures, with every claim cited back to the source text. The goal was answers an analyst can audit in seconds instead of taking on trust.
AI financial-reasoning evaluation
Benchmark machine-generated financial analysis against senior human judgment across equities, credit, multi-asset, and macro. The supporting infrastructure grades model reasoning at scale, with the automated judge validated against human analysts directly. Every output carries its provenance back to source.
Synthetic & alternative-data validation
Generate and stress-test data for regimes where real samples are thin, sensitive, or non-stationary, with explicit checks for leakage and distribution shift. The aim is data a team can train and validate on without quietly importing bias.
Production ML pipelines (MLOps)
Replaced fragile notebook-to-production handoffs with tested, monitored pipelines, including shared feature stores so research and production compute identical features, plus drift and performance alerting. Deployment went from weeks of manual re-writing to a repeatable path the team now runs without me.
Author of Modern Analytics Engineering (2026)
Wrote Modern Analytics Engineering (2026), the book that codifies the calibration and validation methods behind trustworthy quantitative models. It is the written form of how I turn research into systems a desk can actually rely on.
More detail on any of these is available on request. I’m happy to share specifics directly. Get in touch.