Projects

Policy consulting, empirical data work, dashboards, and small agent-assisted coding projects.

OECD consultancy

Demand-side policies for sustainable food choices

Consulting work for the OECD Trade and Agriculture Directorate's Agro-food Trade and Markets division on demand-side policies for sustainable food choices. The work combines policy research, empirical analysis, and applied data tools to study how governments can encourage more environmentally sustainable food consumption across OECD member and accession countries.

My contribution has included policy stocktaking, literature review, classification of policy instruments, and comparative analysis of approaches such as consumer education, voluntary collaboration, regulation, fiscal measures, public procurement, food-based dietary guidelines, and public canteens. The work also examines stakeholder engagement, household acceptability, evidence gaps, and trade-offs with affordability and public health.

On the technical side, I developed retrieval-augmented generation workflows and semantic analysis tools to structure policy information, compute cosine-similarity policy-alignment measures, extract evidence on policy effectiveness and evaluation results, and compare policy texts across strategic dimensions. I also used text embeddings and k-means clustering for semantic policy classification, and carried out econometric analysis to estimate the effect of Denmark's livestock emissions tax announcement on beef consumer prices, using difference-in-differences, synthetic control, and synthetic difference-in-differences methods.

Supporting Food Systems Policy Workstreams: Analytical, Digital and Capacity Development

Current consulting work for the OECD Food Systems Team on analytical, digital, and capacity-building workstreams related to food systems policy. One part of the assignment focuses on a RAG-based meta-evaluation of food loss and waste policies, classifying evaluated policies and synthesising evidence on their effectiveness.

I am also developing an interactive online resource for the OECD Toolkit for Equal Opportunities in Food Systems. The tool will present existing analysis in an accessible format for policy-makers and interested users, with a dashboard, self-assessment features, and a specialised RAG-based chatbot built on the relevant documentation.

A third part of the project focuses on capacity building: preparing materials and teaching a practical course on AI applications to policy analysis for the OECD Food Systems Team. The course covers embeddings, retrieval-augmented generation, agentic coding workflows, MCP, skills, Git basics, and hands-on examples using policy data.

Fun GitHub projects

danish-meat-tax - analysis of whether Denmark's 24 June 2024 Green Tripartite livestock carbon tax announcement changed Danish supermarket meat prices. The project builds monthly product-store panels from dagligepriser.dk price histories, normalizes prices to DKK per kilogram or liter, estimates difference-in-differences, event-study, subgroup, and synthetic DiD models, and includes a beef carbon-price calibration. See the draft paper.
faolex-analysis - analysis of more than 40,000 FAOLEX food legislation records, including demand-side and supply-side classification, abstract-based embeddings, translation caching, cosine-similarity analysis against food-policy strategy queries, time trends, country maps, and a unified interactive dashboard.
canarias-uni-ml - Python pipeline for Canary Islands job postings and Spanish university degree catalogs. It scrapes job sources, normalizes geography and contract fields, builds a degree catalog, gates candidate job-degree pairs, computes text-embedding alignment scores, and stores outputs in CSV and SQLite.
prestaciones-por-empleo - Python pipeline for SEPE monthly unemployment-benefit reports. It detects and downloads new Excel files, processes selected benefit sheets, creates a wide Spanish-language dataset, generates AIReF-style static figures, and builds a local HTML dashboard with filters by sex, age, region, province, and period.
Prestaciones-por-empleo-AI-exposure - Spanish labor-market AI exposure pipeline based on Anthropic occupation exposure data, O*NET descriptions, INE CNO-2011 records, EPA microdata, and SEPE occupation reports. It builds taxonomy-aware CNO4 embeddings with local Ollama, estimates exposure with Random Forest, cosine-weighted, and cosine-nearest methods, aggregates CNO4 predictions to OCUP1 and industry-quarter exposure, scrapes SEPE monthly CNO4 dashboard reports into a long disaggregated dataset, and runs occupation-level growth regressions and event-study analyses for unemployment and contracts.
replication-tracker-jpe-ej-ectj - local-first tracker for JPE, JPE Macroeconomics, JPE Microeconomics, The Economic Journal, and The Econometrics Journal. It pulls paper metadata from OpenAlex, maps papers to Dataverse and Zenodo replication repositories, reads repository files and documentation, classifies data availability, and exports dashboard JSON.
codex-claude-setup - portable backup of local Codex and Claude configuration, including rules, skills, agents, hooks, and plugins, while excluding caches, logs, session histories, and temporary databases.