I'm not a developer. I have no computer science degree. But I built a fully automated, multi-instrument trading system that runs 24/5, sends Telegram alerts, makes AI-assisted decisions, and manages risk — all without me touching it. Here's exactly how.
Most guides assume you already know how to code. They throw Python snippets at you with no context, assume you understand API authentication, and leave you completely on your own when something breaks at 2am.
I wanted to build something real — not a toy backtest, but an actual system running live paper trades across multiple instruments with proper risk controls. And I wanted to do it without hiring a developer or spending years learning to code.
The key insight: You don't need to write code from scratch. You need to understand systems well enough to direct an AI that writes the code for you. That's a completely learnable skill.
Here's exactly what the system runs on:
| Layer | Tool | Purpose |
|---|---|---|
| Broker | Interactive Brokers (IBKR) | Live order execution via API |
| Gateway | IBC + TWS | Automated session management |
| AI Brain | Claude Sonnet + DeepSeek | Trade decisions + fallback logic |
| Database | SQLite | Position tracking, trade history |
| Scheduler | launchd (Mac) | Runs agents on precise schedules |
| Alerts | Telegram Bot | Real-time notifications + approvals |
| Data | Finnhub API | Economic calendar, market events |
The system trades four instruments in paper mode simultaneously:
Every 15 minutes, a monitor script runs and checks all open positions. If something needs attention — a position aging past its limit, consecutive losses, unusual unrealized P&L — it fires an alert to Telegram.
For trade entries, Claude analyzes market conditions, checks the economic calendar (no trading 30 minutes before or after major events like NFP or FOMC), and either executes autonomously or escalates to me via Telegram for approval depending on the risk tier.
Risk tiers matter: Low-risk actions execute automatically. Medium-risk actions require my approval via Telegram button. High-risk actions simulate first, show me the outcome, then ask for approval. Critical actions never execute without explicit human sign-off.
My workflow was simple but disciplined:
That's it. No IDE. No Stack Overflow. No developer. Just clear thinking about what the system should do, and an AI that handles the implementation.
The system is currently in paper trading mode — meaning real market data, real execution logic, but no real money at risk. This is intentional. You prove the edge in paper mode before going live. The infrastructure is built. The next step is validating the strategy across enough trades to have statistical confidence.
You don't need to build a trading system. But the pattern transfers to almost any automation project:
That architecture — autonomous operation with human oversight at the right points — is the foundation of everything I build now.