Algorithmic trading · AI systems

I build the AI that builds trading strategies.

Nearly two decades designing systematic strategies — and the last few months building a framework of AI agents that build them for me. You describe a strategy in one sentence; a purpose-built agent writes it, then proves it by loading it into a live StrategyQuant.

Try it as a sentence  ·  "long when EMA(20) crosses above EMA(50), 30-pip stop"

sqx-builder — agentic build

The frontier of my work

A framework for agentic strategy development

Most of algorithmic trading is still done by hand: wire an indicator, drag a block, run a search, finish the scaffold yourself. Over the past few months I've been building the layer above that — a suite of Claude Code agents and skills that turn plain English into real StrategyQuant artifacts, and a live oracle that verifies every one of them against a running platform before it counts.

The principle is simple: an AI can propose, but nothing is trusted until StrategyQuant itself accepts it. Generation is cheap; verification is everything.

Input

One sentence

A plain-English trading idea — the entry, the exit, the intent.

Agent

sqx-builder

A Sonnet agent resolves every atom against your install and writes a finished strategy.

Oracle

Live SQX load

The .sqx is loaded into a running StrategyQuant over REST. LOAD PASS = accepted.

BUILDING
Judge

Graded & graduated

A whole population is grid-tested and judged on DSR + PBO — graduate, revise, or kill.

The skill suite — natural language in, import-ready artifact out

skill · sqx-strategy

Strategy Builder

"make me a strategy that trades a session breakout with an ATR trail"

A finished, ready-to-backtest .sqx — every rule and parameter concrete. A finished strategy, not a template with holes. Driven by the sqx-builder agent.

skill · sqx-custom-block

Custom Block Builder

"a trailing stop two ATRs under the Hull moving average"

A validated, import-ready AlgoWizard custom block, built from the indicators your own StrategyQuant install actually has.

skill · sqx-strategy-template

Strategy Template Builder

"breakouts only in the trend direction, session-gated"

A ready-to-import builder template — the scaffold the SQX builder fills as it searches — with a real, falsifiable entry thesis.

skill · sqx-random-group

Random Group Builder

"a Value group of EMA, KAMA and ATR periods"

A ready-to-import random group — the curated menu the builder draws one item from per strategy to fill a slot.

The strategy oracle

The piece I'm proudest of. Every strategy the agent builds is loaded into a running StrategyQuant X over its REST API — the platform parses the strategy and either accepts or rejects it. A LOAD PASS is proof, not a promise: bad atoms and unproven shapes get caught here, before you ever waste a backtest on them.

It's now growing from a load oracle into a trade oracle — driving StrategyQuant's retester headlessly to read back real trades and statistics, no clicking required.

$ load EURUSD_EMA_Cross.sqx
  → endpoint algowizard/loadFile
  ✓ LOAD PASS — SQX accepted the XML
 
$ retest --headless
  trades 211 · PF 1.85 · net 4,672
example readout
Measured, not claimed — oracle-gated, 3-pass denoised
30/30
Unseen real-user prompts load into live SQX
~47s
Median build, single-pass, end to end
100/100
Strategies traded in the breadth battery
37%
Agent cost cut, zero correctness lost
Correctness is gated by loading each build into a live StrategyQuant; speed and cost are measured over 3 median-denoised passes with held-out out-of-sample prompts.
What I'm building now

The Template Factory

The next layer ties it all together: take a trading hypothesis, have the skills author the template and its groups, oracle-check that it loads, run a grid of ~10,000 strategies across instruments, and judge the whole population on out-of-sample statistics — Deflated Sharpe and the Probability of Backtest Overfitting. A template earns a verdict: graduate, revise, or kill. Judged as a population, never by its best offspring.

hypothesis author oracle 10k-strategy grid DSR + PBO gate graduate / revise / kill

The rest of the toolkit

Everything else I do

The agentic work sits on nearly two decades of hands-on systematic trading. The fundamentals are still the service.

01

Custom Indicators & Blocks

Bespoke indicators, signals and building blocks for StrategyQuant. 135 already published →

02

Strategy Development

Idea to live system — generation, robustness testing, walk-forward and deployment, with anti-overfitting rigor throughout.

03

Robustness Audits

An independent second opinion on a strategy before you risk capital — Monte Carlo, edge-ratio, walk-forward, PBO/DSR.

04

Machine Learning for Trading

Predictive models, feature engineering and regime classification — with the discipline to know when ML actually helps.

05

Automation & Tooling

Custom databank columns, Monte Carlo methods, and Python automation around StrategyQuant — sqcli pipelines and batch research.

06

Consulting & Mentoring

Hands-on guidance for traders, quants and teams — strategy design, system architecture, and research process.

19 years in quantitative trading StrategyQuant AI architect 135 published Codebase blocks 30/30 oracle-verified builds Claude Code · Python · MT4/5

Let's build something that holds up.

Whether it's a custom block, a full strategy, an AI workflow around StrategyQuant, or an honest second opinion — tell me what you're working on.

Get in touch