Lead Data Scientist · AI Engineer · Professor · Mathematical Engineer
20+ years in quantitative, analytical, and technical roles · United Kingdom & Portugal · Python (typed, NumPy-first)

"Knowledge is knowing a tomato is a fruit; wisdom is not putting it in a fruit salad." — Miles Kington

I build end-to-end machine learning, AI, and data science systems — from data pipelines and statistical modelling through evaluation, deployment, monitoring, and business-facing decision support. The work combines mathematical and statistical depth with production engineering, with particular attention to model validation, uncertainty, interpretability, drift, and operational reliability. Alongside it I run reproducible research pipelines that turn open data into auditable evidence; when a method I need is missing from the Python ecosystem, I build it, validate it against the reference implementation, and publish it.

Delivered outcomes: 80% reduction in reporting costs · 30% reduction in analytics processing time · €500K reduction in inventory value through forecasting and operational optimisation.

I work across research, engineering, and business-facing delivery — translating complex, often messy problems into systems that can be evaluated, explained, deployed, and used to make decisions. Lean models, robust software practice, results you can reproduce and defend.

• Production AI & LLM systems — RAG, agents, MCP servers, structured outputs, and guardrails, with evaluation, observability, and CI from the start.
• Forecasting, anomaly detection & reliability under shift — classical to foundation-model time series, conformal intervals, drift monitoring, and models that abstain instead of guessing.
• Statistical modelling & causal inference — calibration and uncertainty, class imbalance, cost-sensitive thresholds, and honest model selection under real-world noise.
• Econometrics & policy research — panel and causal models, event studies, synthetic control, csQCA/fsQCA, and Monte Carlo over open economic data.
• Optimisation & decision systems — MILP unit commitment, inventory and replenishment policy, scheduling under uncertainty, cost-weighted operating thresholds.
• ML engineering & delivery — data pipelines, real-time analytics, serving, monitoring, and drift detection — applied across sensor and IoT modelling, healthcare analytics, and NLP.
• Research software — methods missing from the Python ecosystem: typed, tested, validated against the reference implementation, and published with a DOI.

→ Full stack, domains, and model families on Methods.

• feedback-intelligence-agent — Production-style RAG: a customer feedback intelligence agent with FastAPI, evaluation, observability, and CI.
• hf-data-agent — Internal data agent where UI, HTTP, MCP, and Slack funnel into one Agent API, grounding an open-source model in a company knowledge base.
• clinic-forecasting-platform — Healthcare demand forecasting: a 13-model benchmark with conformal intervals, rolling-origin backtesting, and FastAPI serving.
• setqca — Native, typed Python csQCA/fsQCA with exact Boolean minimisation — not an R wrapper. Matches the reference R package; on PyPI with a DOI.
• portugal-public-pension-financing — How the public pension promise was actually financed, separating legal obligations, cash accounting, and actuarial liabilities before calling anything a deficit.
• bmssp ⭐ — Deterministic single-source shortest paths via a BMSSP-style divide-and-conquer design (typed, tested).

Live dashboards: Portugal Economic Indicators · NASDAQ Stock Analytics

→ The full catalogue and this year's highlights on Projects.

• Production RAG and agentic systems with evaluation, observability, and CI baked in — MCP as a first-class entrypoint alongside HTTP and Slack
• Portuguese public finance and official statistics under audit: what the general-government balance, the pension promise, and a headline index actually measure
• Reliability under distribution shift: conformal coverage with explicit abstention, survival-model drift monitoring, cost-weighted operating thresholds
• Configurational methods in Python: a native csQCA/fsQCA implementation validated against R, and survey-design-aware applications across the EU-27

→ The full set of active threads on Research.

I teach mathematics and data subjects at ESMAD (Instituto Politécnico do Porto) and run seminars on MLOps, streaming analytics, experimentation, and forecasting — see Teaching.

Open to collaboration on production AI, reproducible policy research, robust time series, and forecast-to-decision systems — see Research. When reaching out, include a short note on your use case, constraints, and timeline so we can assess fit quickly.