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.