← Back to Community

My AI Trading Stack: GPT-4, Python, and Alpaca API

Alex_Quant·
#gpt-4#python#alpaca#setup

Related Link:

https://github.com/example/gpt4-trading-stack →

After 6 months of experimentation, I've landed on a stack that actually works for AI-powered systematic trading.

The Stack

Signal Generation: GPT-4 API
Execution Framework: Python 3.11 + Backtrader
Brokerage: Alpaca API (commission-free, great API)
Data: Polygon.io (real-time + historical)
Deployment: AWS Lambda (serverless, runs every 15 min)

How It Works

  1. Market Scan (GPT-4) — Every 15 minutes, I feed GPT-4 a prompt with:

    • Current price action for my watchlist (50 stocks)
    • Recent news headlines (scraped from FinViz)
    • Technical indicator values (RSI, MACD, volume)

  2. Signal Extraction — GPT-4 returns a structured JSON with trade recommendations:

    {
  "symbol": "AAPL",
  "action": "BUY",
  "confidence": 0.78,
  "reasoning": "Strong volume breakout + bullish news..."
}

  3. Execution — Python script validates the signal (checks confidence > 0.7) and submits orders via Alpaca API.

Results So Far

6-month backtest: +18% (vs SPY +12%)
3-month live trading: +7.2% (small account, $5k)

Not life-changing yet, but it's beating buy-and-hold with zero manual intervention.

Key Learnings

  • GPT-4 hallucinates if you don't constrain the output format. Use strict JSON schemas.
  • News sentiment is gold. Adding news to the prompt boosted win rate from 54% to 61%.
  • Backtests lie. Live results are always worse. Plan for slippage and API latency.

GitHub repo linked above—feedback welcome!