Investment strategies & trading systems
Systematic strategies across equities, FX and futures, from signal research and robust validation to trading systems with pre-trade risk controls, order management and monitoring.
Investment Technology Consulting · Quantitative Finance & AI
One architect for investment technology, from strategy to production.
I help financial institutions design and build investment strategies, trading systems, and the AI, data and analytics platforms behind them. I combine fifteen years inside banks and investment teams with hands-on engineering across statistics, privacy, networks and infrastructure, so the solution works for the business, the risk function and the IT team.
Across the whole stack
- 01Bank operations & governanceControls, approvals, audit, compliance
- 02Investment process & behaviourResearch, portfolios, risk, decisions
- 03Strategies & trading systemsEquities, FX, futures; signals, execution, risk
- 04Statistics & AIModels that hold up out of sample
- 05Data to productionPipelines, quality, deployment, monitoring
- 06Privacy & securityData stays where it belongs
- 07Networks & infrastructureHardware, on-premises and cloud
Services
What I help with
For banks, asset and wealth managers, family offices and fintech teams.
AI solution architecture
AI assistants and agent workflows over financial data, covering retrieval, tool access, evaluation and human approval steps.
Privacy-first AI & infrastructure
Architectures that keep sensitive data in your control: read-only access, self-hosted or private deployment, tenant isolation, encrypted collaboration.
Data platforms & ML to production
From vendor feeds to monitored production systems: ETL, data quality, experiment tracking, deployment, CI and observability.
Portfolio, risk & behavioural analytics
Attribution, factor and non-linear risk models, and analytics that show which investment decisions drove results.
Governance for AI & models
Approval gates, audit trails, validation and monitoring designed the way a bank's risk and compliance functions expect.
Approach
Investment technology usually fails in the hand-offs.
Business, quants, data engineers, IT, security and compliance each own one layer, and value leaks between them: a model that can't be deployed, an AI tool that can't touch the data it needs, a system nobody can audit. I have worked inside each of these layers, so one person can carry a solution from the first conversation to production.
Bank operations & governance
Ten years at J.P. Morgan (asset management and investment bank) and BMO. Designed an attribution methodology adopted firmwide. Technical PM for regulatory-compliant data workflows.
Investment process & behaviour
Fifteen years across research, portfolio construction, risk and attribution, on equity mandates of USD 100M–900M. Systematic equity long/short, FX and futures strategies. Doctoral studies in behavioural finance.
Statistics & AI
Ensemble ML, neural networks, NLP and LLM agents. Non-linear risk models. Holdout, walk-forward and multiple-testing control.
Data to production
ETL and data-quality pipelines, MLflow research platforms, containerised Azure workers, CI, automated tests, LLM tracing.
Privacy & security
Read-only data access for AI agents, end-to-end encrypted AI collaboration (Ponteo.AI), tenant-isolated knowledge stores.
Networks & infrastructure
Proxmox virtualisation, self-hosted GitLab and CI, Docker services behind a reverse proxy, IoT messaging (MQTT), Azure.
Leadership
Led teams of PhDs and developers as Head of Data Science at a Zürich hedge fund. Coordinated IT and solution architects at J.P. Morgan.
What this means
One person who can talk to the portfolio manager, the risk committee and the network engineer, and build what all three need.
How an engagement works
- 01Assess
Understand the process, the data, the constraints and what success means for the business.
- 02Architect
One design across data, models, privacy, governance and infrastructure, with the trade-offs made explicit.
- 03Build
Hands-on delivery of the core system, with tests, monitoring and audit trails from the start.
- 04Hand over
Documentation and knowledge transfer so your team can run and extend it.
Behavioural lens
Systems are used by people, and people have patterns.
My doctoral research was in behavioural finance. It shapes how I design every system.
01
Explain the decision, not just the number. Analytics should show which choices drove an outcome: timing, sizing, holding losers, selling winners.
02
Design against bias. Holdout data, pre-registered limits and multiple-testing control stop confirmation bias from reaching production.
03
Keep humans where judgement matters. AI proposes and challenges. Approval gates and deterministic rules decide what matters.
Product
Alpha · invited testersPonteo.AI
A private room for your agents. Your AI and a partner's AI work on sensitive matters together, so documents stop being shuttled between them.
- Sealed. Every room is end-to-end encrypted. The relay only ever holds ciphertext and metadata.
- Signed. Every item carries its author's signature, and safety codes confirm who is on the other side.
- Under your control. Nothing leaves without your OK. The connector blocks obvious secrets, and the relay runs on your own server, in your country.
- Open to any assistant. Works with any AI assistant that speaks MCP: Claude Code, Cursor, ChatGPT, Codex and others.
Selected work
Systems I have designed and built
Institutional and independent projects, each with the problem, what was built and the value it creates. Institutional work is described at a level that respects confidentiality.
Private AI analyst over portfolio data
Delivered- Problem
- Analysts and clients want to question sensitive portfolio data in plain language, without giving an AI write access or exporting the data.
- Built
- A client-facing AI analytical assistant. Proprietary data is exposed read-only through a self-hosted MCP server. Retrieval combines a knowledge graph with a vector store, analysis comes with automated charts, and role-specific modes include a portfolio-manager view. Web and terminal interfaces.
- Value
- Natural-language access to proprietary analytics, with the firm in control of what the AI can see and do.
- AI architecture
- Privacy
- RAG
Investment strategy research: equities, FX, futures
Delivered- Problem
- Most backtests flatter. Strategies need to be designed and validated the way a capital allocator would assess them.
- Built
-
- A USD 50M systematic equity long/short strategy anchored by a behaviour-aware short-signal model (2Xideas).
- Hybrid alpha engines for global equity mandates (J.P. Morgan, BMO).
- Systematic FX strategies and an ensemble FX direction classifier (Trius Partners).
- Intraday equity-index futures strategies with walk-forward evaluation, bootstrapped pass probabilities, stress tests and a human-approved execution bridge, plus trend-following futures research (independent).
- Value
- Research comes with an honest estimate of how likely it is to hold up, not a single attractive equity curve.
- Strategies
- Statistics
- Validation
Governed agentic trading system
In development- Problem
- AI agents can speed up research, but no risk function should let them control capital.
- Built
- Agents propose, review and monitor, and deterministic code decides. Components:
- pre-trade risk engine with portfolio-level limits and a kill switch;
- order state machine with reconciliation;
- lifecycle with a one-time holdout and human approval;
- hash-chained audit log;
- offline test suite run against simulated brokers.
- Value
- A multi-asset trading system (FX, gold, crypto) in which AI speeds up research while risk, approvals and execution stay under deterministic control.
- Trading systems
- Governance
- AI agents
Behavioural analytics platform, built from scratch
Delivered- Problem
- Standard attribution says what a portfolio returned, not which decisions drove it. A new firm needed the full platform to deliver that insight.
- Built
- As sole quant, data engineer and developer: a proprietary behavioural attribution model, portfolio reconstruction, multi-period attribution linking, a data-quality and corporate-action pipeline, a columnar market-data lake, and containerised cloud workers on Azure.
- Value
- Research turned into a production platform by one person, from model design to cloud deployment.
- Data to production
- Behavioural
- Cloud
Multi-tenant knowledge platform for AI agents
Prototype- Problem
- Teams want agents with long-term memory, but one client's knowledge must never leak into another's.
- Built
- A knowledge-graph platform with an isolated namespace per tenant, plus agent-to-agent services in which crews of specialist agents query typed slices of the graph through an MCP server. Anonymised insights are distilled into a shared layer.
- Value
- Shared learning across clients without sharing anyone's data.
- Privacy
- AI agents
- Knowledge graphs
Self-hosted infrastructure and edge systems
Running- Problem
- Some data cannot leave the building, so AI and analytics have to run where the data lives.
- Built
- Proxmox virtualisation hosting a private GitLab with CI, containerised services behind a reverse proxy, and LLM tracing for agent observability. At the University of Zürich: a microservices architecture for IoT bioreactors (MQTT, PostgreSQL, Grafana).
- Value
- On-premises, edge or cloud, chosen on the merits rather than tooling limits.
- Infrastructure
- Networks
- Hardware
Risk, attribution and research platforms
Delivered- Problem
- Institutional portfolios need analytics that portfolio managers trust and control functions accept.
- Built
-
- A residual-based Days-on-Risk attribution model, adopted firmwide for performance attribution and capital budgeting.
- A bespoke non-linear equity risk model.
- Regression-based attribution engines and a high-fidelity backtest framework.
- An ML research platform with experiment tracking and data governance, led as Head of Data Science.
- Value
- Quantitative methods embedded in institutional processes, not left in research decks.
- Governance
- Analytics
- Data to production
No projects match this filter.
About
Graziano Mirata
More than fifteen years in quantitative finance on the buy side and sell side: J.P. Morgan Asset Management, BMO Asset Management, J.P. Morgan Corporate & Investment Bank, the Zürich hedge fund 2Xideas (Head of Data Science, later Head of Systematic Strategies), Trius Partners and Yainvest.ai. Roles spanned portfolio management, quantitative research, data science leadership and hands-on engineering.
I started Quantitative Intelligence Studio in April 2026 to bring that combination to financial institutions as an independent consultant.
- Experience15+ years, buy side and sell side, London and Zürich
- DomainsAsset management, investment banking, hedge funds, wealth management, fintech
- EducationDoctoral studies in Behavioral Finance and MSc in Finance, University of Lugano (USI)
- LanguagesEnglish, Italian (native) · German (basic)
- Based inSchwyz, Switzerland
Contact
Have a system to design, build or rescue? Let's talk.
- Emailgraziano.mirata@quantiativeintelligencestudio.com
- LinkedInlinkedin.com/in/g-mirata
- LocationSchwyz, Switzerland