Operação Serenata de Amor — Open-Source AI Auditing Brazilian Federal Deputies' Expense Claims

Operação Serenata de Amor was launched on 7 September 2016 by data scientist Irio Musskopf, sociologist Eduardo Cuducos and entrepreneur Felipe Cabral, with crowdfunding on Catarse that reached 131% of its goal and financed three months of development. It is a civil-society initiative (now under Open Knowledge Brasil), not a government system.
Its robot 'Rosie' is an open-source Python application that ingests receipts from Brazil's Chamber of Deputies Quota for the Exercise of Parliamentary Activity (CEAP), enriches them with open data (Revenue Service company records, Google, Foursquare, Yelp) and uses unsupervised learning to estimate a 'probability of corruption' per reimbursement, listing the reasons it was flagged.
Reported results: over 3 million financial documents analysed, roughly 8,000 suspicious cases identified, and 629 complaints against 216 federal deputies filed with the Chamber's Ombudsman by early 2017. Some politicians publicly apologised and promised repayment; the one documented repayment was about BRL 700.
Limits are plain: the tool covers mainly the Chamber (clearer written rules than the Senate), flags are statistical suspicions rather than proof, measurable recovered funds were small, and the project relies on volunteers and recurring donations of roughly BRL 10,000 a month. Its value lies in transparent, replicable citizen oversight of public spending.

Read the full analysis: https://d3.harvard.edu/platform-rctom/submission/operation-love-serenade-fighting-corruption-in-brazil-with-an-open-source-machine-learning-powered-robot/