ShadowDesk

Autonomous ML-Powered Trading Platform with Docker Compose local deployment.

Overview

ShadowDesk is a system-agnostic autonomous trading platform that provides:

• Real-time stock signal generation via ML inference
• Automated trade execution
• Full ML pipeline (training, validation, deployment)
• Observability stack (metrics, logging, dashboards)
• Demo mode with paper trading (no real money)

⚠️ Security Notice

This is an educational and portfolio demonstration project.

Local Development Credentials

The docker-compose.yml contains default development credentials for local services (MinIO, Vault, Grafana, Airflow, etc.):

• Intentionally included for local development only
• Not production credentials - for Docker Compose local stack only
• Not connected to real trading - uses Alpaca paper trading

Production Secrets

Production secrets are managed via:

• HashiCorp Vault - Primary secrets management
• Sealed Secrets - GitOps-safe encrypted secrets for Kubernetes

Never commit real API keys or production credentials.

Architecture

ShadowDesk/
├── docker-compose.yml     # Local deployment (dev credentials only)
├── config/               # Local configs (prometheus.yml, kind-config.yaml)
├── docker-config/         # Docker build configs
├── project/               # Source code
│   ├── cluster_b/
│   │   ├── api/         # FastAPI trading service
│   │   ├── train/       # ML training & Airflow DAGs
│   │   └── trading/     # Trading execution (Alpaca broker)
│   ├── dashboard-streamlit/  # Streamlit dashboard
│   ├── shared/          # MinIO, Vault, Qdrant helpers
│   └── tests/           # Test suite + local validation
├── k8s/                  # Kubernetes manifests + Helm charts
│   ├── argocd/          # ArgoCD AppProject & Applications
│   ├── helm/            # shadowdesk-infra, shadowdesk-compute
│   └── sealed-secrets/  # Encrypted secrets (safe to commit)
└── docs/                 # Documentation

Local Quick Start

Prerequisites

• Docker and Docker Compose
• 8GB+ RAM recommended

Start the Stack

docker-compose up -d
docker-compose ps
docker-compose logs -f fastapi

Seed Local Data

cd project
python tests/local_stack_bootstrap.py

Service URLs

Service	URL	Credentials
FastAPI	http://localhost:18000	-
Streamlit	http://localhost:18501	admin/shadowdesk
Airflow	http://localhost:18080	admin / [generated at runtime]
MLflow	http://localhost:15000	-
Prometheus	http://localhost:19090	-
Grafana	http://localhost:13000	admin/admin
Vault	http://localhost:18200	dev token: shadowdesk-devtoken
Qdrant	http://localhost:16333	-
MinIO Console	http://localhost:19001	shadowdesk/shadowdesk123

Retrieve Airflow password:

docker compose exec -T airflow cat /opt/airflow/simple_auth_manager_passwords.json.generated

Development

Code Quality (Required for PRs)

ruff check .
black --check .
isort --check-only .
pytest tests/ -v

Install Dev Tools

pip install ruff black isort pytest pytest-cov

Run Tests

cd project
pytest tests/ -v

Validate Local Stack

python tests/validate_local_stack.py --bootstrap  # Full validation

API Endpoints

Account

• GET /account/ - Account info
• GET /account/portfolio - Portfolio summary
• GET /account/history - Trade history

Trading

• GET /trading/signals/{symbol} - Get signal
• POST /trading/signals/batch - Batch signals
• GET /trading/positions - Current positions
• POST /trading/orders - Submit order
• DELETE /trading/orders/{order_id} - Cancel order
• POST /trading/cycle - Run full trading cycle

Demo Mode

When no Alpaca API keys are configured, the system runs in demo mode:

• Mock $100,000 portfolio
• Simulated positions (AAPL, NVDA, MSFT)
• Heuristic signals when model unavailable

CI/CD

GitOps pipeline: GitHub Actions → ArgoCD → Kubernetes

Stage	Tools
Lint	ruff, black, isort
Test	pytest, pytest-cov
Deploy	ArgoCD CLI, Helm

License

MIT - see LICENSE