Initiative file / Infrastructure
dIAra: AI-based public infrastructure monitoring
Evidence record v2
41 / 42
- Institution
- CGR
- Editorial classification
- Verified adoption
- Initiative type
- AI system
- Documented phase
- Pilot
- Relationship to AI in the source
- Explicit relationship to AI
- Execution evidence
- Confirmed
- First evidence
- Last verified
- Next review: Nov 29, 2026
Documented scope
What it is
CGR pilot with Universidad Cenfotec and German cooperation agency GIZ combining IoT image capture with AI detection of people and machinery. Code and technical reports document Faster R-CNN with ResNet50 and inference on real images from a construction site in Pococí. Verification is limited to that experience, not generalized operation or current continuity.
Objective stated by the institution
Support public-works monitoring and oversight through images, indicators of people and machinery presence, and potential alerts for human review and citizen participation.
Known timeline
Only milestones with an explicit date in the sources or editorial review are shown.
- First evidence
- Latest editorial verification
- Next scheduled review
Evidence reading
Each dimension is assessed separately. Undetermined marks an evidence gap; not applicable means the dimension does not correspond to the documented object.
What is confirmed
- The repository contains code, training, architecture documentation and inference and CSV examples produced on real images from the Pococí site. The first located evidence is the snapshot published on May 10, 2024, not a confirmed initial deployment date.
- The team itself warns that the model was trained for that site and its effectiveness may change on other construction projects. The OGP submission separately identifies an application still needed to automate execution and reporting.
What could not be determined
- Current inference continuity, AI coverage by site, end-to-end automation, a consolidated total budget and independent oversight outcomes were not established. The pilot phase describes the latest substantiated technical experience, not a guarantee that it remains active today.
Documented results
Results are published only when a source traceably attributes them to this initiative.
The team reports mean IoU of 82.7% and recall of 85.4% for the documented model. These are technical metrics from a specific construction-site experience, not an independent evaluation or evidence of detected irregularities or recovered public funds.
Evidence matrix
Sources and traceability
Sources state which claims they support, their origin and the date on which they were consulted.
- [1]CGR dIAra repository: May 2024 technical snapshot
- Publisher
- Contraloría General de la República / Universidad Cenfotec
- Source type
- Official primary source
- Published
- Consulted
Supports: existence · execution · AI technique · reported result
- [2]Detection system documentation: Faster R-CNN with ResNet50
- Publisher
- Contraloría General de la República / Universidad Cenfotec
- Source type
- Official primary source
- Consulted
Supports: AI technique · execution
- [3]AI-stage results: inference and output files on construction images
- Publisher
- Contraloría General de la República / Universidad Cenfotec
- Source type
- Official primary source
- Consulted
Supports: execution · AI technique · operational use · reported result
- [4]CGR submission to OGP: AI alerts for citizen control of infrastructure
- Publisher
- Open Government Partnership, propuesta de la CGR
- Source type
- Multilateral organization
- Consulted
Supports: existence · stated objective · target · governance
Open questions
- At which sites and dates were AI models actually run, and which still have active inference?
- Was the automation application completed, and what controls, out-of-sample validation and oversight outcomes were documented?
Context
This October 2026 addition concerns development and tests published in 2024, not a launch this month. Detecting the person class does not identify individuals or establish facial recognition. It is not presented as a corruption or irregularity detection system, and AI use is not attributed to every site receiving devices. It is distinct from CGR's internal Audit planning tool.
Related projects
Ecosystem and capabilitiesOperational
Verified:
LINC: Community Innovation Labs
MICITT network of Community Innovation Labs for training, co-creation and local solution development. Its offering includes programming, digital fabrication, robotics, cybersecurity, artificial intelligence and other emerging technologies.
- Initiative type
- Capacity-building program
- Operational start
Ecosystem and capabilitiesIn development
Verified:
CONECTA: national interoperability based on X-Road
National Digital Government Agency project to build a roadmap and adopt a national interoperability model based on X-Road. Thirty institutions participated in the initial diagnostic exercise.
- Initiative type
- Digital infrastructure
- Announcement
Ecosystem and capabilitiesPlanned
Verified:
LaNIA: National AI Laboratory
National Artificial Intelligence Laboratory proposal presented by CeNAT at the start of national-strategy formulation in February 2023. Its stated objectives include training, awareness and development of high-impact initiatives.
- Initiative type
- Capacity-building program
- Announcement