What I learned connecting AI signals to real trading accounts

I built an AI trading executor and here’s what I learned connecting AI signals to real trading accounts

I’ve been working on IMALI, an AI-assisted trading platform designed to analyze markets, score trading opportunities, and execute trades through a user’s existing exchange/broker account.

One of the biggest challenges wasn’t generating signals. It was everything that comes after the signal:

• deciding whether a signal is actually strong enough to trade
• position sizing and risk controls
• handling API failures and exchange differences
• tracking open positions
• preventing duplicate or stale trades
• separating the trading engine from custody of user funds

I ultimately went with a non-custodial model. The user keeps their funds at their supported trading platform, while IMALI connects through API credentials and handles the analysis/execution layer.

I’m curious how other developers here approach this.

If you were building an automated trading system today, would you rather build your own execution engine from scratch or connect an AI strategy layer to existing exchanges/brokers?

I’m also interested in feedback on the architecture and risk controls as I continue improving the platform.