Portfolio

Selected work

Projects from forecasting, supply chains, healthcare, public policy, pricing, and recommendation systems.

A mix of industry work, research, and open-source projects.

Case study

01

Development economics / Applied NLP

I designed a research prototype that turns English and Arabic news into location-level evidence for food-security analysts across Iraq, Jordan, Lebanon, Palestine, and Syria.

The system matches 167 risk factors across 12 themes and resolves mentions against 357 geographic entities. Analysts can trace every flag back to its article, source, language, and location.

Signal strength and evidence confidence are reported separately. The output is a queue for investigation, with a plan for future outcome validation.

Read the case study
Illustrative analyst verification queue showing separate signal and confidence measures
Illustrative verification output. The prototype prioritizes analyst review; these scores do not establish crisis probabilities.

Industry work

02

Causal Foundry / Applied data science

Demand forecasting, procurement, and health financing

I build predictive models and decision-support systems across national medical supply chains, health financing, and facility operations.

  • Build demand-forecasting models and procurement recommendations to support supply planning.
  • Lead the capitation payment methodology and analyse facility behaviour to inform payment design and resource allocation.
  • Develop analytical pipelines over more than 100M claims, with data representing over 9M insured people.
  • Build automated indicators and anomaly detection across 1,000+ providers, including utilization, prescribing quality, and performance monitoring.
  • Reconcile more than 5,000 drug SKUs across seven agencies using text similarity and embeddings.
Read the capitation reform paper

ZF Group / Commercial ML

Pricing leakage detection and optimization

I developed anomaly-detection, elasticity, uplift, and optimization models to support commercial pricing decisions.

  • Pricing-arbitrage detection that protected approximately €11M in annual leakage.
  • Models across a €500M+ product portfolio, contributing to a 13% gross-profit improvement.
  • PySpark and Databricks pipelines developed with engineering partners.

Benshi AI / Digital health

Engagement, forecasting, and adaptive interventions

I worked on forecasting and experimentation for pharmacy supply chains, healthcare workers, and mobile-health applications.

  • Demand forecasts that helped reduce pharmacy stockouts by 18%.
  • Survival models and churn prediction to understand engagement.
  • Adaptive interventions using contextual bandits and reinforcement-learning concepts.
Read the KDD engagement paper

Research & code

03

Experimentation / Decision-making

Deep Contextual Bandits

Experimentation workflows built around Bayesian neural networks, Thompson sampling, and decisions under uncertainty, adapting ideas from the Deep Bayesian Bandits benchmark.

Source code

NLP / Research

Text as evidence

Work on retrieval, classification, similarity, and network analysis for public-policy and customer-behavior questions, including newspaper-based research on political violence at IAE-CSIC.

Background and experience

Background and research