dIAra — Costa Rica's Auditor Trains Cameras and a Neural Network on Public Construction Sites

Costa Rica's Contraloría General de la República (CGR), the country's supreme audit institution, built dIAra (Dispositivo de Inteligencia Artificial para Reconocimientos y Alertas) to monitor public construction projects in real time, aiming to deter irregularities, raise citizen trust, and give auditors earlier warning of problems on-site.
The system pairs low-cost, solar-powered cameras — built from open-source designs and deployable on a MiFi connection — with a neural network trained on hundreds of site photographs to recognise construction machinery and people. Cameras capture images at set intervals and upload them to the cloud, where the model estimates how much work and equipment activity is happening, flagging projects whose declared progress diverges from what the images show so auditors can prioritise a closer look.
The CGR has published the camera hardware plans and the AI models openly on its GitHub, explicitly so that any government body — in Costa Rica or elsewhere — can replicate the build, from constructing the cameras to applying the models. As of 2025 the CGR had two complete mobile dIAra kits (two to three cameras, a MiFi router and a solar panel each), moved between sites, with a 2025 target of covering at least ten construction projects; a pilot at a school under construction was reported as completed with 'very good results,' though no quantified before/after figures on deviations caught or costs avoided have been published.
The project was submitted to the Open Government Partnership's Open Gov Challenge as 'Artificial Intelligence Alerts for Citizen Control of Public Infrastructure,' and the CGR presented it to fellow Latin American supreme audit institutions through OLACEFS (the Latin American and Caribbean organisation of supreme audit institutions) via a dedicated capacity-building webinar, positioning it as a model other national audit offices could adopt directly from the open repository.
Caveat: this is still an early-stage pilot — only two camera kits and a single fully reported deployment (the school project) as of the latest public reporting — so its evidence of prevented corruption or cost savings is qualitative, not yet backed by published statistics, unlike more mature systems in the region.

Read the full analysis: https://algoritmos.uniandes.edu.co/diara-dispositivo-de-inteligencia-artificial-para-reconocimientos-y-alertas/