Skip to content
Back to work

CASE STUDIES

A trading bot built to fail safely

An automated trading system with the engine separated from the screen, an enforced risk layer, and a hundred-odd tests for the days the market does not cooperate.

Year
2026
Role
Design and engineering, end to end
Client
Personal

Code that places real orders fails in ways ordinary software does not: duplicate entries, state that disagrees with the exchange, or a clock a few seconds out that gets every request rejected.

The Python engine talks to the dashboard through a shared database, so it can restart without losing state. Entries are idempotent, position state is reconciled from fills that actually matched, and the stop-loss guard runs on every tick rather than only at entry.

The strategy cleared a two-year backtest and an out-of-sample window before it was allowed to run live across several symbols at once.