Built and hardened Interactive Brokers REST gateway lifecycle and order lookup paths.
Lumibot
Trading infrastructure earns trust at its boundaries: broker state, time, data, and failure recovery.
Role
I maintain broker and runtime paths, build AI-agent infrastructure, improve performance, and strengthen release evidence across backtest, paper, and live execution.
Constraint
Brokers disagree about lifecycle and state. A path can look correct in isolation while failing at startup, returning stale orders, or diverging between a backtest and a live account.
Selected work
Strengthened broker startup, import, clock, and paper-trading gates around real strategy runs.
Worked on an agent runtime with financial tools, structured memory, replayable traces, and explicit trading permissions.
Made representative backtests 30–40× faster while lowering median RAM use by roughly 80%.
Evidence
2,000+ GitHub stars on the public project.
71 merged pull requests authored by Martin as of August 2026.
Performance results published on Martin's LinkedIn profile.
Inspect source