STF's Victor OCR processes 10 million+ pages; classifier pending

AI system

System: Victor (STF–UnB artificial-intelligence project: OCR, document splitter, document classifier, repercussão geral theme classifier)
Model: Not assessed
Release: Not assessed
Application developer: Supremo Tribunal Federal (STF) with the Universidade de Brasília (UnB)
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: STF–UnB research partnership with about 40 graduate scholarship holders. Building a training set of extraordinary appeals, which took months. Automatic OCR of all incoming appeal PDFs on the day of receipt (from December 2020). Validation of any theme indication by the ministers.
Human oversight: Not assessed
Relation to AI: By the STF's own account of 19 August 2021, the AI classifier that would flag repercussão geral themes was not in production. The only live component was automatic OCR of incoming appeals, which the STF frames as part of the Victor project. Any theme indication is validated by the ministers. The time savings announced in 2018 (three hours or 40 minutes to five seconds) were forecasts made by chief justices at launch. All cited sources are the deployer's news service (STF) and the CNJ, which the same chief justice presided. No independent evaluation was found.
AI contribution: By the STF's own account of 19 August 2021, the AI classifier that would flag repercussão geral themes was not in production. The only live component was automatic OCR of incoming appeals, which the STF frames as part of the Victor project. Any theme indication is validated by the ministers. The time savings announced in 2018 (three hours or 40 minutes to five seconds) were forecasts made by chief justices at launch. All cited sources are the deployer's news service (STF) and the CNJ, which the same chief justice presided. No independent evaluation was found.

Benefits

Outcome identifiers: Not assessed
Benefit: more than 10 million pages processed by May 2021
Category: Repetitive Task Automation
Status: Not assessed
Beneficiary groups: People. Organizations. Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Measurement

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: more than 10 million pages processed by May 2021
Comparison: Assumed alternative: a court clerk converting image files to text by hand
Results: Assumed comparison. Reference: três horas. Result: cinco segundos.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Money

Kind of value: Not assessed
Amount: Not assessed
Currency and prices: Not assessed
Period and unit: Not assessed
Realization: Not assessed
Recipients and payers: Not assessed
Gross or net: Not assessed
Included costs: Not assessed
Comparison: Not assessed
Method: Not assessed
Return and payback: Not assessed

People and places

Participants: Developer: Supremo Tribunal Federal (STF) with the Universidade de Brasília (UnB). Operator: Supremo Tribunal Federal (STF), Secretaria de Gestão de Precedentes.
Jurisdictions: Benefit: Brasília, Brazil. Developer: Brazil. Operator: Brazil.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Follow-up

Conditions for replication: Filings arriving as PDFs that can be converted to text. Splitter and classifier reaching production (not achieved by August 2021). Classifier accuracy acceptable to the Secretaria de Gestão de Precedentes. Continued STF–UnB funding and staff.
Replication attempts: 1 follow-up entries. Outcome unresolved: 1.
Later findings: 11 May 2023. In May 2023 the STF said it still operated Victor for repercussão geral triage of incoming appeals, alongside Rafa, while finishing tests of the new VitórIA tool.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Evidence

Source numbers: AIBD-0333-source-0001–0005
Citations: Numbered references below.
Language: Not assessed
Source relationships: Developer, Independent
Original evidence: Not assessed
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Grades

Assessed benefits: People. Organizations. Society.
Depth: Not assessed
Reasons: People: No litigant outcome is documented. Organizations: OCR of 10 million pages is a throughput figure, not a measured saving, and the classifier was not in production. Society: No effect on case backlog or litigants is documented.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently studied

Entry details

Kind: Not assessed
Related entries: Not assessed
Field: Justice administration: appellate case triage. Supreme court case management.
Purpose: Speed up admissibility analysis of extraordinary appeals by flagging the repercussão geral theme that applies. The production OCR layer is a precondition for that, not the classification itself.
Dates: Date basis unassessed: 2021 (precision unassessed). Updated: 21 October 2026.
Stage: Not assessed
Version: SHA-256: 4d5f8f71116f…

References

Other identifiers: Not assessed
Category mappings: Not assessed
Brief: Not assessed
Revisions: 5 revisions. Latest: 21 October 2026.

Sources

Report Independent author

Montenegro, M. C. (2018). Inteligência artificial: Trabalho judicial de 40 minutos pode ser feito em 5 segundos. Agência CNJ de Notícias, 23 Oct 2018.
Press release Developer

Supremo Tribunal Federal (2021). Projeto Victor avança em pesquisa e desenvolvimento para identificação dos temas de repercussão geral. STF Notícias, 19 Aug 2021.
Press release Developer

Supremo Tribunal Federal (2018). Ministra Cármen Lúcia anuncia início de funcionamento do Projeto Victor, de inteligência artificial. STF Notícias, 30 Aug 2018.
Press release Developer

Supremo Tribunal Federal (2023). STF finaliza testes de nova ferramenta de Inteligência Artificial. STF news, 11 May 2023.