Dynasty Engine — Multi-Asset RL Trading Platform

Overview

Dynasty Engine autonomously trades equities, crypto futures & ETFs using deep reinforcement-learning with strict risk controls. It ships real-time telemetry via Prometheus/Grafana plus a Next.js dashboard & Expo iOS app for operations.

Spring 2025 Update:

• Core modules now feature comprehensive error handling, retry logic, and robust logging for all critical paths (RL policy, risk manager, execution, asset manager, backtest harness).
• Prometheus metrics endpoint is now fully integrated with the Flask API and Cloud Run via Gunicorn.
• Dockerfile updated to use Gunicorn for production API serving.
• All changes backward-compatible and reviewed for safety.

Features

• Interactive Brokers multi-asset execution
• Deep RL policy (Stable-Baselines3 PPO) with Optuna hyper-param search
• Kelly sizing & draw-down risk manager
• Robust error handling and retry logic throughout all core modules
• Prometheus metrics endpoint integrated with Flask API (Gunicorn/Cloud Run ready)
• Prometheus metrics exporter + Expo push alerts
• Next.js dashboard (lp-portal) & Expo mobile app (apps/dynasty-mobile)
• Pre-commit (Black, Ruff) & GitHub Actions CI (tests, lint, nightly retrain)
• Docker Compose for local dev & monitoring stack

Deployment

1. Install deps

   poetry install                    # Python core
   npm --prefix lp-portal i          # Web portal
   npm --prefix apps/dynasty-mobile i # Mobile

2. Copy env files & fill secrets (POLYGON_KEY, TRADIER_KEY, SUPABASE_* …).

3. Local all-in-one stack

   docker compose up            # engine + Prom + Grafana + portal
   # http://localhost:8000/metrics , http://localhost:3000 , http://localhost:3001

4. Manual run (no Docker)

   python metrics_exporter.py &
   python multi_asset_manager.py --symbol AAPL
   npm --prefix lp-portal run dev
   expo start -p 8081

5. Production – push to main; CI pipeline runs tests, lint, retrains model, and can deploy to Cloud Run.

Secrets

• IBKR API credentials, OpenAI API key fetched securely from GCP Secret Manager

Monitoring

• Prometheus scrape endpoint :8000/metrics (now fully integrated with Flask API, Gunicorn, and Cloud Run)

Key metrics:

• dynasty_drawdown_pct
• dynasty_action_total{symbol,action}
• dynasty_model_version{symbol}
• dynasty_rl_infer_ms{symbol}

Visualised in Grafana dashboard (grafana/dynasty.json). Push alerts delivered via Expo.

Prometheus

Run python metrics_exporter.py or use docker compose which already includes Prometheus.

Back‑test

python backtest_harness.py MSTR 2022-01-01 2024-01-01 data/MSTR.csv

Tests

pytest -q

---

Robustness & Reliability Improvements (Spring 2025)

• All critical Python modules (trading_env.py, multi_asset_manager.py, rl_policy.py, risk_manager.py, execution.py, backtest_harness.py) now feature comprehensive error handling, retry logic, and robust logging.
• Prometheus metrics endpoint is now directly integrated with the Flask API and Gunicorn for Cloud Run compatibility.
• Dockerfile and requirements have been cleaned and updated for production reliability.
• These changes were made to maximize reliability and observability without breaking downstream API or analytics.

Author

• Esh, Founder of Entropie LLC