FridayTrader

Autonomous AI trading agent using Claude + RSI, MACD, and volume analysis to make real-time buy/sell decisions on NASDAQ stocks. Runs twice daily with a live dashboard tracking P&L vs S&P 500.

Comment

Inspiration

I wanted to learn how AI could be applied to real-world financial decisions. Most trading bots just use simple rules — I wanted to build one that could reason like a human trader, explaining why it makes each decision.

What it does

FridayTrader is a fully autonomous paper trading agent that analyzes NASDAQ stocks twice daily and makes BUY/SELL/HOLD decisions. It calculates RSI, MACD, volume ratio, and 5-day trend, then sends the full technical context + live news headlines to Claude (Anthropic API) for a final decision. Every trade is logged with Claude's reasoning in Obsidian markdown files.

How I built it

  • Data layer: yfinance fetches 60 days of price/volume history
  • Indicators: RSI, MACD, volume ratio, and 5-day trend calculated from scratch in numpy
  • AI layer: Claude (claude-sonnet-4-6) receives a structured prompt with all technical signals and makes the trading decision
  • Execution: Automatic stop-loss (5%) and take-profit (15%)
  • Dashboard: Flask server with live portfolio heatmap, allocation bars, and P&L vs S&P 500
  • Scheduler: Runs automatically Mon-Fri at market open and midday SGT

Challenges

The hardest part was getting Claude to follow strict trading rules consistently. Early versions would buy stocks even when RSI was overbought. I solved this by pre-scoring each stock with a technical score before passing it to Claude, giving it clear signal context.

What I learned

How to combine traditional technical analysis with LLM reasoning, build a live Flask dashboard, and deploy a fully autonomous agent that runs without human intervention.

Built With

Share this project:

Updates