A literature observatory built on local language models: what the literature is about, how research is done and where it is produced. The models run on your own computer through Ollama, so no text leaves your machine.
Details ↓
Collects articles from SciELO and OpenAlex, maps topics over time with embeddings, classifies each abstract against a codebook you write (quoting the passage behind each answer, and validated against blind human coding), and counts production by state, country and institution. Runs from the command line, from Python or from a local web panel, and publishes as a static site. The pilot covers ten Brazilian political science journals, 2010–2025. Successor to SciELO-Summarizer (SICSS Brasil 2024).
One of the most complete databases on Brazilian municipalities, organized in 17 thematic dimensions and documented as an R package. Made in partnership with researchers at MAPE, with whom I build and maintain the database and its public interface.
The public monitoring platform of the RESA project, covering deforestation in Acre and in the Chico Mendes Extractive Reserve.
Details ↓
Maps PRODES/INPE deforestation data for Acre and the Chico Mendes Extractive Reserve, year by year, and compares the reserve with its surroundings. Part of the RESA project, on what agroextractive cooperatives change in household income and deforestation in Acre, selected in an Instituto Clima e Sociedade call supported by the Bezos Earth Fund. I coordinate the project’s quantitative work, under the overall coordination of Luci Maria Teston (UFAC).
Who is related to whom in Brazilian politics. Language models read the entries of the Dicionário Histórico-Biográfico Brasileiro (DHBB) and record each family tie they state, and every link in the network cites the passage it came from.
Comida di Buteco is a contest that picks the best bars in cities across Brazil. This project uses web scraping and geolocation to map those bars nationwide.
The programme of the 15th ABCP Meeting (Belém, 2026) in real time: what’s on right now, a personal agenda, a calendar, and search across every paper and abstract. What you mark stays on your device.
A dashboard built on the mape_municipios infrastructure for the office of state deputy Dani Balbi, bringing together MUNIC, PNS, and Censo SUAS data on the LGBTQIA+ population of the state of Rio de Janeiro, including the territorial distribution of social assistance facilities.
Joint work with Tomás Paixão Borges (IESP-UERJ).
private deployment
Python
Client & lab work
Delivered for organizations and research groups; not publicly available.
2025 · FGV EBAPE
Digitized and systematized historical election results with OCR and LLMs.
2024–25 · CEBRAP
OCR and LLMs applied to the Brazilian Statistical Yearbooks to systematize social assistance data since the early 20th century.
2024–25 · GENI/UFF
LLM and RAG tools for structured extraction from unstructured documents.
2024–25 · Fundação Roberto Marinho
Dashboards in Streamlit and Power BI, and an AWS data lake.
2024 · FGV
A RAG pipeline for querying FGV researchers’ publications in natural language.
2022–25 · Instituto Rio21
An automated Python pipeline that cut report production from days to about an hour, and interactive maps for corporate audiences.
