I build data and AI projects on top of public data, and I treat every result as something a reader should be able to check. Methodology comes first and gets written down before the results, a finding that refutes the hypothesis is reported anyway, and verification scripts ship with the code.
| Project | What it answers |
|---|---|
| Raio-X Patrimonial · live demo ↗ | Electoral history and declared assets of Brazil's 513 federal representatives, with gender and race breakdowns. Data matched against the Electoral Court (TSE) for 511/513 records |
| Capital Allocators Backtest | Monthly contributions from 2006 to 2025, seen by a Brazilian investor: capital allocators vs. ETFs vs. fixed income, in real terms |
| Atribuição de Canais | Multi-touch attribution with cross-device identity, seven models with error margins, and an incrementality test, under Brazil's data-protection law (LGPD) |
| micropolis-agent | Why LLM agents play SimCity badly: is it spatial reasoning, or just representation and memory? |
| Radar Educação Superior (in progress) | Where higher-education supply grows without turning seats into students and graduates |
Portuguese (native) · English (professional) · Spanish and German (basic)
Construo projetos de dados e IA sobre dados públicos, e todo resultado precisa poder ser conferido por quem lê. A metodologia vem antes e fica registrada antes dos resultados, um achado que contraria a hipótese aparece do mesmo jeito, e o script de verificação vai junto com o código. Os projetos estão documentados em português.