I design, attack and defend AI systems in regulated banking environments.
role: AI Security Architect @ BBVA Technology
focus: AI Safety | Red Teaming | Adversarial ML | Financial Crime
proof: 206+ attack vectors evaluated, 7 critical findings pre-deploy
ranking: Kaggle Master | HackTheBox Top 800 Global (L4tentNoise)
education: BS Computational Mathematics & CS (AI) | MSc Gen AI & Deep Learning
location: Madrid, Spain | Open to EU/UK remote| Where | What | Result |
|---|---|---|
| BBVA | AI Red Teaming on LLM/RAG/Agent systems in FinCrime | 206+ vectors 7 critical -20% latency -35% costs |
| BBVA | Secure RAG + Fraud Detection in production banking | +22% AUC +15% precision 500K+ guarded queries/mo |
| Ecoembes | Edge CV waste classification + NLP assistant | 85% accuracy 45K img/h <100ms -18% CO2 |
| Capgemini | AWS data lake + forecasting + BI automation | +20% forecast -30% cycle 10h/analyst/wk saved |
MITRE ATLAS OWASP LLM Top 10 PyRIT Garak Red Teaming Threat Modeling
PyTorch Transformers LangGraph LangChain RAG GraphRAG FAISS
DGX TensorRT-LLM Triton NIM CUDA FP8/INT8/AWQ
Kubernetes Docker AWS Azure MLflow Zero Trust CI/CD
Python C++ TypeScript SQL
| Degree | Focus |
|---|---|
| BSc Applied Mathematics & Computing | AI, Cryptography, Bayesian Statistics, Optimization |
| MSc Deep Learning & Generative AI (240h) | Neural Networks, NLP, GenAI |
| MSc Data Science & Big Data (240h) | ML pipelines, Big Data, Analytics |
| MSc Smart Solutions IoT (240h) | Edge computing, IoT systems |
| Project | Stack | What it proves |
|---|---|---|
| Spectra | LangGraph + Neo4j | AI attack surface recon with graph reasoning |
| FraudAI Agent | LangGraph + RAG + Qdrant | 6-agent fraud platform, 482 tests, 93% coverage |
| Hospital Center AI | LangGraph + FastAPI + PostgreSQL | 8 parallel medical agents, consensus routing |
| Drone GeoAnalysis | YOLOv11 + Flask + Mapbox | ISR platform with LLM mission planning |
| WatchDogs OSINT | LangGraph + GPT Vision | 4 parallel CV agents, score 95/100 |
| Triagegeist | XGBoost + CatBoost + SHAP | Kaggle: 99.98% triage prediction accuracy |