Geneva & Vienna · open to data & AI roles
I build RAG and LLM systems, data pipelines and quantitative evidence for international organisations and industry. An anthropologist turned data scientist, I work where machine learning meets public policy and human behaviour.
About
Most data problems are people problems in disguise. I combine production-minded engineering with the methods of a policy analyst and the curiosity of an anthropologist, so the systems I build answer the question that actually matters. I work end to end, from scraping and ETL to models, dashboards and the briefing that turns them into decisions.
Retrieval-augmented generation, graph and vector databases, embeddings and automated reporting, shipped in Docker on Azure and AWS. I use agentic coding tools such as Claude Code to move from prototype to working system quickly.
Impact evaluation, regression and comparative analysis, time-series forecasting and procurement analytics. Trained at CEU in Vienna, applied with the World Bank, the OAS and the EU.
An anthropology degree, first in class, and fieldwork on armed conflict, elections and migration. It taught me that evidence should come from both data and people's testimonies, and to explain results to engineers, doctors and diplomats alike.
Selected work
A selection of projects across AI engineering, data science and policy research. Client work is described without confidential names or data.
An automated reporting engine, not a chatbot. From a standard template and user inputs it pulls news and API data, builds a fact and relationship graph of events, people and organisations, and drafts reports grounded in traceable sources. Shipped with a containerised UI for Azure.
Outcome: used in crisis reporting, including the 2026 Venezuela earthquake; country offices trained on RAG-based reporting.
Read the case study →End-to-end system that scrapes, cleans and structures public procurement data at scale, then feeds a RAG and graph knowledge base for competitor reporting, BI dashboards and pricing and bidding models inside a custom CRM, for manufacturing and security-sector clients.
Scale: 1M+ contracts and 8k+ competitors structured.
Read the case study →A monthly panel built from National Police records and ACLED events for Buenaventura, Arauca and Tumaco, estimated with two-way fixed effects and event studies, plus robustness checks and QGIS maps.
Finding: the onset of peace talks is linked to more monthly violence, led by political violence.
Code on GitHub →With Paula Biša: a contract-level Corruption Risk Index from red-flag indicators validated by logistic regression, comparing the Latvian implementer of Rail Baltica with the rest of the national market on TED data, 2018 to 2023.
Finding: no sign of unusually high risk overall; the implementer sits within, and partly below, market norms.
Code & deck on GitHub →Seeded and unseeded LDA, DBSCAN and a Spanish transformer sentiment model on 43 presidential speeches and ~8,800 Reddit sentences. Builds on earlier NLP work at the OAS in Santo Domingo.
10.6 pp more negative affect in migration sentences on Reddit, 4.6 pp in speeches.
Code on GitHub →Led the CEU workstream: cleaned and merged international TIMSS data and ran survey-weighted regression and comparative analysis of socio-economic and parental-investment gaps.
ETL pipelines on Copernicus and Sentinel data feeding ARIMA, XGBoost and NARX time-series models, with Dash, Plotly and Power BI dashboards for monitoring.
Machine learning and deep learning linking geographic and environmental attributes to disease progression, presented to doctors, engineers and designers.
25% improvement over baseline findings.
k-means clustering shows how voting patterns cross municipal boundaries. Peer-reviewed article with Milanese.
Census and CaliEnDatos data, regression and neighbourhood clustering turned into a policy brief with three costed options on employment and education gaps for Afro-Colombian communities.
Brief on GitHub →◆ Code samples and notebooks live on GitHub.
Experience
From research labs in Cali to multilateral institutions and private-sector clients, often working in parallel with full-time study.
RAG and graph knowledge base, scraping and ETL over 1M+ federal procurement contracts, BI dashboards, pricing and bidding ML and Apollo API enrichment in a custom CRM. Data architecture from ERDs to user flows.
Built an automated, graph-grounded AI reporting system with a Docker UI for Azure; trained country offices on RAG and LLM reporting; supported needs analysis and mapping of affected populations during earthquake responses.
Led the CEU research workstream, liaised with World Bank counterparts and ran regression-based comparative analysis on TIMSS data.
Dashboards for data storytelling, satellite-based forecasting models and GeoJSON datasets for radio coverage, environmental monitoring and infrastructure planning.
Reorganised 20 years of cooperation archives (about 8 tons, 9,300 files), built Excel automation and met 100% of the processing goal.
ML and deep learning on geospatial health data; owned the analysis pipeline from data quality to reproducibility.
Observed the 2024 Salvadoran elections with a legal and gender focus; NLP research on migration discourse and support to high-level field missions.
Security and peacebuilding research, geospatial dashboards in QGIS and Power BI, and co-authored policy papers.
Toolkit
Python and SQL every day, R for econometrics, and a cloud and container stack to put things in front of users.
Education
Central European University, Vienna. Impact evaluation, text analysis and NLP, procurement analytics, data visualisation.
Universidad Icesi, Cali. First in class, Dean's List. Thesis on civilian and guerrilla interaction in Urabá.
University of Turin, Italy.
Languages
Recognition & publications
FLACSO Ecuador, 2nd edition.
Milanese & Navarro (2025), Revista Geographia. Plus contributions to SegLab policy papers on security, peace and the pandemic.
MEAL · Satellite image processing · Project planning & evaluation in health research
Contact
Hiring for a data scientist, AI or data engineer in Geneva, Lausanne, Vienna or remote? I'm happy to talk about roles, projects and collaborations. I can work in Austria without visa sponsorship.