Level 7 · 60–80 h

Production, MLOps, regulation, career and capstone

What turns research into a professional system: maintainable software, monitored infrastructure, models with a life cycle, the legal and tax framework, the career and a defended final project.

Fig. — Maximum CFD leverage for retail clients in the EU (ESMA)
08/08
Prerequisite
Levels 4 to 6
Profile
Quant who wants a professional system

Software engineering for trading

  • 7.1.1 Project structure and configuration (Pydantic Settings, YAML); Ruff, pre-commit, pytest, tests with simulated brokers.
  • 7.1.2 Performance: profiling (the Python 3.15 sampling profiler), Numba, Cython; Rust (PyO3) as background.
  • 7.1.3 Patterns: event-driven, message queues (Redis, ZeroMQ, Kafka), hexagonal architecture.

Infrastructure and deployment

  • 7.2.1 Docker and docker-compose; reproducible images with uv.
  • 7.2.2 VPS or cloud (AWS, GCP, Azure, Hetzner) close to the broker's or exchange's servers; CI/CD with GitHub Actions.
  • 7.2.3 Orchestration (cron, Prefect, Airflow, Dagster) and monitoring (structured logs, Prometheus and Grafana, alerts, heartbeats).
  • 7.2.4 Security and continuity: secrets, 2FA, IP allowlists, backups, procedures for broker, data or network failures.

MLOps for trading models

  • 7.3.1 Experiment and model tracking (MLflow, Weights & Biases); data versioning (DVC).
  • 7.3.2 Scheduled retraining, automatic validation and champion/challenger deployment.
  • 7.3.3 Monitoring drift and live performance; automatic deactivation.
  • 7.3.4 Serving models: ONNX, CPU or GPU inference, the cost of LLMs in production.

Regulation, ethics and taxation

General information for Spain and the EU, with a global view. It is not advice.

  • 7.4.1 MiFID II: client classification (retail or professional), suitability and appropriateness; algorithmic trading requirements for firms (RTS 6) as background.
  • 7.4.2 ESMA and CFDs: maximum leverage for retail clients (30:1 on major currency pairs; 20:1 on non-major currencies, gold and major indices; 10:1 on other commodities and indices; 5:1 on equities; 2:1 on crypto), position close-out at 50% of margin and negative balance protection.
  • 7.4.3 PRIIPs and KID: why an EU retail investor cannot buy US-domiciled ETFs; alternatives (UCITS, options, professional client status).
  • 7.4.4 MiCA: the transitional period in Spain ended on 1 July 2026; trade only with authorised or passported CASPs; checking the CNMV and ESMA registers.
  • 7.4.5 Market abuse (MAR): spoofing, layering, manipulation; why a badly designed bot can engage in them.
  • 7.4.6 Taxation in Spain: IRPF (Spanish personal income tax) savings base (19% up to €6,000; 21% up to €50,000; 23% up to €200,000; 27% up to €300,000; 30% above that), the two-month rule for repurchases, loss offsetting (4 years; 25% cross-offset limit), FIFO for crypto, Modelo 720 and 721 (foreign asset and crypto declarations), withholding at source and double taxation treaties.
  • 7.4.7 Ethics: personal data and GDPR, scraping, responsible AI, conflicts of interest in selling “signals”.
Fig. — Spanish income tax rates on the savings base by bracket (common territory)

Quant career

  • 7.5.1 Roles: quant researcher, quant developer, quant trader, data engineer, risk quant, ML engineer.
  • 7.5.2 Where to work: hedge funds, prop firms, banks, asset managers, market makers, crypto, fintech; the Spanish and European ecosystem.
  • 7.5.3 Public portfolio (GitHub, blog, Numerai, Kaggle) and interviews (probability, statistics, coding, brainteasers).
  • 7.5.4 Certifications and continuing education (CQF, EPAT, CFA, FRM, master's degrees); independent trading and prop firms, and their risks.

Final capstone

40 to 80 hours. Choose one of these projects and defend it before a panel:

  • C1 Multi-asset futures system (trend, carry and spreads) in NautilusTrader with paper trading at IBKR.
  • C2 Systematic options book (premium selling with tail hedges), volatility surface and real-time Greeks management.
  • C3 Equity and ETF portfolio with ML (cross-sectional ranking with LightGBM, HRP and rebalancing with costs and Spanish taxation).
  • C4 Research platform with LLMs and alternative data (RAG, text signals and an analyst agent with guardrails).
  • C5 Crypto market making or arbitrage with ccxt, funding and basis, and counterparty risk control.

Common requirements: reproducible repository (uv and Docker), documented data, anti-overfitting validation (walk-forward and CPCV or Deflated Sharpe), risk report, at least four weeks of paper trading, monitoring and technical report.

Research on this site to practise this level

Practice

  1. The Working Environment - 1Research-0013 · 4 min