I built an algorithmic trading system, and the most important part is not any individual strategy — it is the layer that decides whether any strategy gets to trade at all.
Nine independent strategies run in parallel across MT5, each generating signals. Before any position opens, they go to an LLM veto layer. The LLM reads the current signal, checks it against live market context, and decides: does this trade make sense right now?
If not, it blocks the trade.
The infrastructure also includes:
- A Master/Slave trade copier built in MQL5 for multi-account execution
- SL/TP management and position sizing logic
- A feedback loop logging outcomes back into model context over time
- Multi-asset support: BTC, ETH, and FX pairs
Building it taught me more about system architecture under real financial pressure than any other project I have worked on.
Interested in this project?
I'm always happy to talk through how it was built, the problems it solves, or how something similar could work for you.