Julio César Abril Medina is CTO at GSE – Gestión de Seguridad Electrónica S.A., a Colombian company with over 16 years of experience in digital identity, biometrics, and digital trust. A Systems and Computer Engineer (UPTC), he holds a Master's degree in DevOps & Cloud Computing (Universitat de Catalunya) and a Continuing Education Master's in Artificial Intelligence: Model Management and Deployment (UCAM, Universidad Católica San Antonio de Murcia). He is an ISO 22301 Internal Auditor, certified in ISO/IEC 42001 (AI Management System), and also holds TOGAF and AWS Architecting on AWS certifications. With over 12 years of experience in cloud security architecture, PKI, and digital identity, he leads multi-account security governance on AWS, HSM/KMS-based identity strategy, and the integration of facial biometrics with identity and access management (IAM) in projects such as the one described in this presentation.

Frictionless Identity: LLMs Applied to Biometric and Access Event Analysis in the Herta–BeMyself Integration AbstractPhysical access control and workforce identity management today generate a growing volume of events: facial verifications, spoofing attempts, logins, and user provisioning or deprovisioning. These events are typically scattered across technical logs that only a specialist can interpret, delaying incident detection and complicating auditing.Herta provides the identification layer through BioAccess, its proprietary facial recognition engine: a 1:N identification algorithm capable of performing up to 100 million comparisons per second, with liveness detection (anti-spoofing) and over 90% accuracy even with face masks, glasses, or partial facial occlusions. BeMyself provides the workforce identity (IAM) layer: Single Sign-On, passwordless authentication, and automated user provisioning, generating its own stream of logical access events. At the center of this talk is the intelligence layer that connects both worlds: a large language model (LLM) that ingests events from Herta and BeMyself, correlates facial identifications with logical access, detects anomalous patterns, and answers natural-language questions such as who attempted access outside business hours or which users have accumulated failed spoofing attempts. The LLM turns raw biometric and IAM telemetry into human-readable explanations and audit reports for security teams, without requiring manual dashboard or log queries. The talk will also cover the risks of introducing an LLM over sensitive biometric data and the AI governance measures applied, aligned with ISO/IEC 42001, to ensure traceability, explainability, and human oversight over decisions. It will close with use cases in corporate, financial, and industrial environments, and lessons learned from implementing this converged identity architecture.