RecomendeMe Intelligence: AI and Knowledge Graphs Against Human Trafficking
Human trafficking and child exploitation have reached a critical point, with a 25% increase in victim detection in 2024 according to UNODC data. In Brazil, this complexity is magnified by regional inequality and multi-city trafficking corridors. To address these challenges, RecomendeMe Intelligence — the investigative research arm of the RecomendeMe platform, led by Lucas Matheus — has developed a framework that utilizes Artificial Intelligence (AI) and Knowledge Graphs to convert fragmented public data into actionable intelligence.
The 4-Layer Framework: From Ingestion to Synthesis
RecomendeMe's technology addresses four critical investigative "failure modes": overwhelming data volume, the use of coded language, information siloing, and vulnerability to leaks. The architecture is divided into four fundamental stages:
1. Ingest and Linguistic Decoding
Utilizing tools like spaCy and Grok AI (xAI), the system processes heterogeneous data from OSINT sources — such as Federal Revenue records (CNPJ/CPF), transparency portals, and judicial journals. A key differentiator is the automated detection of coded keywords and trafficking-adjacent euphemisms used by recruiters to evade traditional filters.
2. Behavioral Logic and Modus Operandi (MO) Profiling
Extracted entities are scored against a behavioral pattern library derived from UNODC case studies and public court records. This allows the system to identify criminal networks not just by names, but by repetitive suspicious behaviors.
3. Network Structuralization (Neo4j)
Validated data is exported to a Neo4j Graph Database, transforming isolated data points into dynamic nodes and relationships. This stage visualizes the infrastructure of criminal networks in Brazil, exposing hidden hubs and local connections across various jurisdictions.
4. Criminological Synthesis (Human-in-the-Loop)
To ensure methodological rigor and legal admissibility, every finding must pass through a mandatory Human-in-the-Loop (HITL) checkpoint involving criminology experts before any escalation to authorities or the media.
Actionable Results: The Epstein Case in Brazil
The system's efficacy was proven during an investigation conducted between October 2025 and March 2026. The platform processed tens of thousands of pages from the "Epstein files" in under 48 hours to reveal the first actionable signals of connections within the country.
RecomendeMe Intelligence mapped exploitation network connections across eight Brazilian cities:
- São Paulo
- Rio de Janeiro
- Natal
- Brasília
- Vitória
- Fortaleza
- Belo Horizonte
- Santa Catarina
The structured evidence was formally submitted to the Brazilian Federal Public Prosecutor's Office (MPF) in February 2026 (Dispatch No. 5/2026/UNIC/SC1PGR), resulting in an active official investigation. Furthermore, the data supported investigative reporting by BBC News Brasil and other independent outlets.
Ethics, Transparency, and the Future of Civic Intelligence
The platform operates strictly with public-domain data, ensuring full compliance with Brazil's LGPD and ITU ethical transparency principles. By democratizing access to tools that previously required massive institutional resources, RecomendeMe promotes a decentralized model of civic intelligence.
Next Steps
- Financial Intelligence — Integration of public financial flow signals (CNPJ) to detect monetary ties between network nodes.
- Scope Expansion — Adapting ontologies to combat labor trafficking and domestic servitude.
- Open Source — Plans to release the investigative toolkit (NLP ontologies and Neo4j schemas) as open-source to empower NGOs and journalists worldwide.
RecomendeMe Intelligence demonstrates that by uniting AI precision with human curation and the transparency of knowledge graphs, it is possible to illuminate the shadows where the most complex exploitation networks operate.