Applied AI engineer with 18 years in capital markets technology. I build LLM agents and developer tools that take over real work in trading and QA teams, then test them like the production systems they run beside.
Hover a field to see what it means. Explaining FIX traffic in plain language, and spotting when it's wrong, is what my testing agents at Tradeweb do all day.
Most of my AI work replaces something a team used to buy or do by hand. These are the projects I'm proudest of.
Agents generate and run FIX scenarios against client order and execution flows, then explain each failure in plain language so a tester doesn't have to read raw tag-value logs.
LLM agents · FIX 4.x · Python
Turns business requirements into structured, reviewable test cases. Testers start from a solid first draft instead of a blank page.
LLMs · prompt design · QA workflow integration
An in-house test-management system with AI built in, used across the QA organization in place of a commercial licence.
Python · web app · LLM features
Checks trading-system logs against expected behaviour and flags anomalies worth a human look, so issues surface before clients notice them.
LLMs · log analysis · Linux
Backtests of AI trading agents can't be trusted: the model has already read about what happened. This grades agents only on calls sealed before the outcome, using a commit-reveal hash chain that anyone can verify in their browser.
Python · MCP server · Claude Code plugin · GitHub Actions
Analyst, bull and bear, trader and portfolio-manager agents debate each stock and commit to a call. A nightly news scan triggers reviews of earlier calls when something material changes, and every call feeds the scorecard above.
Claude Code subagents · Python data toolkit · scheduled jobs
Answered support and code questions for the Options Vol trading desk, built before most teams had started using LLMs.
Python · OpenAI · LangChain
Eighteen years on fixed income, rates and risk systems at banks, hedge funds and market operators.
The tools change every quarter. These are the parts that stay useful.
Multi-agent orchestration, tool use and MCP, prompt design, Claude and OpenAI APIs, LangChain.
Evaluation, calibration and guardrails, from a career spent breaking trading systems before clients could.
FIX order and execution flow, fixed income, MBS and rates, market data, risk and P&L.
Python, Java, SQL, FastAPI, CI/CD, Docker, Kubernetes, AWS, Linux.