Brazil’s open-source AI auditor for congressional expenses

What it does

Serenata de Amor is a Brazilian civic-tech initiative that applies data science to public administration. Its main component, Rosie, is an artificial intelligence that analyzes expenses from Brazil’s Chamber of Deputies, Federal Senate, and municipalities to surface suspicious patterns. The findings are served through Jarbas, a web interface where citizens inspect receipts and suspicions, while Rosie also engages directly with politicians on Twitter.

The interesting bit

The project treats government oversight as a data pipeline: raw spending data goes in, anomalies come out, and public pressure follows. What makes it unusual is the deliberate pairing of automated detection with human-scale accountability—an algorithm that doesn’t just crunch numbers but names names in public.

Key highlights

• Rosie audits expenses from Brazil’s Chamber of Deputies, Federal Senate, and municipalities.
• Jarbas runs continuously, serving expenditure data and suspicion details around the clock.
• The team maintains a separate serenata-toolbox package for dataset generation and reuse.
• The project is explicitly designed for any country with open spending data, though Brazil is the primary focus.

Caveats

• Updates are infrequent: the README states the project has expanded into new efforts and core infrastructure now runs manually only about once a month.
• The maintainers acknowledge the technical stack is legacy—“not the best stack ever”—and direct new contributors seeking active work to the Querido Diário project.

Verdict

Worth exploring if you care about civic tech, government transparency, or applied ML in the public interest. Skip it if you need an actively maintained codebase to build on today.