About us
CyberAntifraud is the umbrella for my PhD research (NOVA IMS) into merchant‑acquiring fraud prevention. The focus is to build robust deep learning methods to:
- Detect sophisticated fraud patterns with strict temporal validation and drift control.
- Reduce merchant friction and fraud loss by optimising the Precision–Recall trade‑off.
- Simulate adversarial tactics with LLMs for robustness testing (fraud red‑teaming).
- Operationalise pipelines across SAS/Oracle and Python/ML stacks.
Note: no sensitive data is disclosed; examples use synthetic or anonymised data.
Facts & Figures
1 TB
Transactional data (Oracle/SAS)
5+
DL families evaluated (RNN/Transformer/MoE/Autoencoder/GNN)
3
Target papers (Survey • Methods • Results)
Q1
Survey publication target
Regulations and Activity reports
The project tracks PSD2/PSD3, RTS SCA, EBA Guidelines, GDPR, NIS2 and reports from professional bodies (e.g., ACFE). We produce internal briefings on ML impact (data, explainability, retention, audit).
- Internal briefings: [placeholders: links/PDFs]
- Reg‑to‑ML mapping: requirements → data & metrics [placeholder]
Research team
| Name | Role | Contact |
|---|
| Paulo Saramago |
Doctoral Researcher • Fraud & Deep Learning (Acquiring) |
paulo@cyberantifraud.com |
| [Supervisor] |
Scientific supervision |
[email] |
| [Co‑supervisor] |
Methods/Infra |
[email] |
Governance
- Steering: supervisors + fraud & compliance stakeholders.
- Reviews: quarterly checkpoints (metrics, drift, error cost).
- Change control: data/model versioning (MLflow/DVC).
Scientific Advisory Board
- [Name • Affiliation • Role]
- [Name • Affiliation • Role]
Research Management and Development
- Data management: field catalog (MCC, device, geo, 3DS, etc.), PII segregation.
- Validation: blocked time‑series CV, rolling windows, ensembles & MoE.
- Ops: SAS ↔ Python pipelines, testing, observability, feature store.
Equality, Culture, Diversity, and Inclusion
Commitment to non‑discrimination and bias assessment in models (auditability & fairness).
Research Ethics
Principles: data minimisation, legitimate anti‑fraud purpose, operational explainability,
explicit error‑cost evaluation (customer vs. risk), full documentation for audit.
Sustainable Development
Reduce payment friction and fraud loss (economic impact) and improve green compute efficiency.
Work with us
Open to academic/industry collaboration on:
Dataset shift & Drift
MoE for Fraud
GNN on merchant networks
LLM Red Teaming
Real‑time Explainability
Contact: collab@cyberantifraud.com
