Trading Bot

An event-driven trading bot in Java 21. It streams live market data, runs a PPO reinforcement-learning agent in-process via ONNX Runtime, places risk checked paper trades on Alpaca, and writes telemetry to TimescaleDB for Grafana. The same strategy code runs both live and in the backtester.

Made while reading Building Trading Bots Using Java - Shekhar Varshney

What it does

• Runs the PPO policy inside the JVM through ONNX Runtime. A golden test checks the Java feature pipeline against the Python training pipeline to 1e-5.
• Talks over a single in-memory event bus, so one Strategy implementation works live and in backtest.
• Sizes orders and runs per-symbol, gross-exposure, and buying-power checks plus a daily-loss kill switch before anything reaches the broker. Positions are reconciled from the broker on startup.
• Only trades when the PPO agent is above a configurable confidence threshold.
• Backtests with commissions and slippage against a buy-and-hold benchmark, prints the metrics to the console, and writes the equity curve and trades to CSV.

Quick start

Run the full stack with Docker:

cp .env.example .env         
cd observability
docker compose up             # bot + TimescaleDB + Grafana on http://localhost:3000

Or run it locally:

cp .env.example .env
mvn spring-boot:run           # live paper session
mvn -B verify                 # build + tests + coverage gate

Backtest (writes backtest-results/equity.csv and trades.csv):

mvn spring-boot:run -Dspring-boot.run.profiles=backtest \
  -Dspring-boot.run.arguments="--strategy=fade --symbols=AAPL,MSFT --days=30 --commission=0.01 --slippage-bps=10"

Tech Stack

Java 21, Spring Boot 3.4, ONNX Runtime, Alpaca, TimescaleDB, Grafana, JUnit 5, Mockito, JaCoCo, GitHub Actions.