Welcome to my personal workspace. I bridge the gap between high-level enterprise strategy and hands-on AI engineering. With over 25 years of experience leading global media operations for industry giants like Disney and Fox, I specialize in transforming legacy creative supply chains into high-margin, cloud-native and local inference deployment ecosystems.
As an AI Expert at Microluma, I spearhead the integration of intelligence layers into existing scientific hardware. This software-defined approach unlocks advanced machine learning capabilities from current technical baselines, democratizing access to high-precision automation.
Visit Microluma.euAs a self-employed consultant, I architect sovereign, decentralized knowledge systems. A primary focus is my work on federated pathology inference clusters, orchestrating local nodes across distributed NVIDIA Blackwell hardware using Mixture of Experts (MoE) architectures and secure Tailscale networking. I also develop advanced open-source audio description pipelines leveraging Visual Language Models (VLMs) and voice synthesis.
End-to-end design and production deployment of AI tools, including Local Inference Clusters, Prompt Engineering, and Retrieval-Augmented Generation (RAG).
Orchestrating multi-region hybrid cloud environments across AWS and Microsoft Azure, alongside distributed compute clusters.
Hands-on coding in Python and TypeScript/React, utilizing Domain Driven Design (DDD) for API pipeline orchestration.
Human Centric & Domain Design Driving
Decades of experience in P&L management, margin optimization, and cross-functional global team leadership.
"Knowledge precedes Intelligence."
An open scientific initiative and computational framework establishing the theoretical, mathematical, and architectural foundations of Knowledge-Centric Clinical Intelligence. Organizes computational medicine around the lifecycle, persistence, and institutional sovereignty of validated clinical knowledge rather than transient statistical models or dataset curation.
A decentralized knowledge aggregation system for pathology utilizing MedGemma as a foundational model. Built for privacy, utilizing local inference nodes and specialized hardware.
View Project DetailsFine-Tuned LLMs • Deterministic Optimization • Zero-Trust Security • Real-Time Disruption RAG
An enterprise-grade decision-support and routing engine engineered for global supply chain resilience and mission-critical reliability. The platform integrates a domain fine-tuned language model governed by strict deterministic mathematical constraints, guaranteeing zero hallucinations in freight allocation, contractual commitments, and scheduling.
Cybersecurity & Zero-Trust Architecture: Built on a strict Zero-Trust security model featuring mutual TLS (mTLS) encrypted telemetry across all ERP and carrier integration endpoints, least-privilege tokenized authentication, and cryptographic tenant isolation—ensuring sensitive shipping manifests, proprietary rate matrices, and route telemetry remain impervious to data leakage or cross-contamination within vector stores.
Predictive Disruption & Multi-Modal Optimization: Leveraging real-time vector search (RAG), the system autonomously monitors global external disruption feeds—including extreme weather events, geopolitical conflicts, maritime chokepoints, and natural disasters—to forecast delay cascades before they occur. An intelligent shipment selector dynamically optimizes and scores multi-modal carrier routes balancing transit latency, carbon intensity, and operational cost.
Vision Language Models • Neural Voice Synthesis • Subtitle QC & Live Broadcasting
A broadcast-grade, end-to-end media accessibility platform built upon 25+ years of global media operations leadership. AT Studio orchestrates local vision-language models (VLMs) and neural speech synthesis to deliver studio-compliant Audio Description (AD), automated Subtitle Quality Control (EBU-TT / SMPTE-TT compliance), and real-time live subtitling—ensuring unreleased creative IP never leaves sovereign local infrastructure.
Audio-Driven Facial Animation • Decoupled Co-Speech Kinematics • Spanish Lip-Sync GAN Harness
An advanced multi-modal digital human framework engineered for high-fidelity, emotion-responsive speech synchronization and natural physical presence. The system departs from conventional phonetic mouth-flapping by utilizing a decoupled two-tier architecture: a high-resolution facial articulation model driven by acoustic prosody and emotional inflection, paired with a co-speech body motion model that synthesizes natural upper-body and hand kinetics.
Spanish Speech-to-Face Foundation & Training Harness: Developed alongside a parallel engineering pipeline for custom Spanish acoustic phonetics, featuring per-frame MediaPipe landmark tracking with dual-window temporal smoothing (eliminating fixed-crop drift), automated dropout recovery, and objective cross-modal validation using SyncNet audio-visual correlation scoring.
My approach to building intelligent systems is deeply rooted in my early education. I hold a Bachelor's Degree in Physics from Universidad Complutense de Madrid, which provided the foundational mathematical framework I use for complex machine learning architectures today. Most recently, I formalized my advanced deep-learning expertise by completing the Generative AI Engineering Fellowship Master's Program at Outskill.