I built an AI trading system that monitors markets, makes decisions, and executes trades — without me touching anything. No coding background. No finance degree. Just a Mac Mini, some Python scripts I built with AI assistance, and a clear process.
Here is exactly how the system works.
Every 15 minutes, my system pulls live market data across four instruments: Bitcoin, Nasdaq futures, Gold futures, and EUR/USD. It analyzes price action, momentum, volatility, and market regime. It decides whether to enter a trade, manage an existing position, or do nothing.
When it fires a signal, it executes the trade instantly through Interactive Brokers — no human in the loop. I get a Telegram alert on my phone for every trade. Then it goes right back to monitoring.
The system has completed 47 trades. I reviewed Telegram alerts every morning. That was it.
This is not a cloud-heavy, subscription-heavy setup. The entire system runs on a Mac Mini M2 sitting in my house:
No n8n. No Zapier. No monthly SaaS fees stacking up. The automation backbone is launchd — macOS's built-in scheduler — running 12 agents that never stop.
The system doesn't just look at price. It classifies each instrument into a market regime — trending, corrective, or ranging — and adjusts its strategy accordingly.
In a trending regime, it looks for breakout entries. In a corrective regime, it looks for mean reversion. In a ranging regime, it mostly stays flat. This regime-awareness is the core of why the system doesn't just trade noise.
On top of that, every check evaluates:
If the conditions aren't right, the system logs a no-trade reason and moves on. Most checks result in no trade. That's by design.
Every Sunday at 8 AM, the system runs a research loop. It pulls the week's trade data, calculates win rates and P&L by instrument, and sends it to Claude for analysis.
Claude reads the performance data and generates improvement proposals — specific, actionable changes to strategy parameters, filters, or logic. Those proposals get routed to a governance layer. I review them in Telegram with one-tap approve, simulate, or defer buttons.
If I approve a proposal, an executor agent implements the code change autonomously. The system literally improves itself week over week, with me as the final decision-maker but not the one doing the work.
Running a live trading system means you need to know when something breaks. My monitor runs every 15 minutes and checks:
Every alert comes through Telegram with context — current price, unrealized P&L, stop distance — and buttons to close the position or snooze the alert directly from my phone.
I started too broad. My first version tried to trade too many instruments at once. The data was thin, the signals were noisy, and I couldn't tell what was working. Now the rule is: prove one instrument before adding another.
I ignored data quality. Early trades were logging $0.00 P&L because fills weren't being captured in time. Always verify your logging before trusting your results.
I over-tweaked on small samples. 13 trades is not enough data to redraw a strategy around. The system now flags when sample sizes are too small to act on.
The system is in paper trading mode — proving edge before risking real capital. 47 trades completed. The loss skew on Nasdaq futures is the current priority — average loss is larger than average win, which means the stop and trailing logic needs work. That's exactly what the Sunday research loop is designed to solve.
Every script in this system was built through AI-assisted development. I described what I needed, Claude wrote the code, I ran it, reported the output, and we iterated. That cycle — describe, run, report, iterate — is the entire workflow.
I've never opened a code editor in the traditional sense. I work entirely in Terminal. I don't touch files manually. Every change is scripted and reversible.
The barrier to building systems like this is lower than it has ever been. You don't need to know Python. You need to know what you want the system to do, and you need the patience to test it properly.
I document this system and others like it at automatededges.com. Subscribe below for new guides every time we publish.
No code required. Practical workflows you can set up this week — sent every time we publish.
Subscribe Free →