Description

Senior AI Engineer en Quadrivia, donde lidero el trabajo que hace que nuestros agentes de voz para healthcare sean fiables frente a pacientes reales. He reconstruido el orquestador conversacional reduciendo ~1s de latencia por turno, creando un patrón de subagentes que detectan red flags clínicos en tiempo real, diseñado un detector de contestadores que pasó del 50% al >93% de detección, y actualmente construyo Starfish, una plataforma de testing donde agentes IA simulan pacientes para hacer stress-testing de agentes en producción.

Antes cofundé Amrap, un entrenador personal IA en tiempo real: app iOS voice-first + backend multi-agente que generaba rutinas personalizadas y entrenaba en el gimnasio. Diseñé la orquestación combinando LLM reasoning con validadores deterministas, reduciendo alucinaciones un 95% con latencias de voz <2s. Llevamos el producto de cero a usuarios de pago.

Previamente Cloud Architect en Google (2,5 años), cuentas enterprise, cerrando contratos multi-millón € y creando soluciones cloud a escala.

Ayudo a empresas a llevar agentes IA de demo a producción: arquitectura, voice/conversational AI, sistemas multi-agente, eval frameworks, fiabilidad y latencia. Disponible para proyectos fractional (1-2 días/semana) y consultas técnicas puntuales.

Industry field of expertise

Languages

Spanish
Native or bilingual
French
Fluent
English
Native or bilingual

Workplace preferences

Can work on-site

Madrid (up to 50km)

Amrap
Founder, Tech Lead
HEALTH AND WELLNESS
November 2024 - February 2026 (1 year and 3 months)
New York, NY, USA
Cofounded Amrap, a real-time AI personal trainer that builds personalized strength routines and coaches users live on the gym floor through voice and video.

- Designed and built the end-to-end product: voice-first iOS app and a multi-agent backend that generates high-personalization routines and coaches workouts in real time, calling out weights and reps, teaching technique, and answering exercise questions on the spot.
- Architected an agent orchestration layer combining LLM reasoning with deterministic and LLM-based validators, cutting hallucinations by 95% and engineered patterns so the remaining 5% wouldn’t degrade the UX.
- Tuned the system for in-gym latency: sub-2-second voice responses during coaching and full routine generation in 30 to 60 seconds, with on-the-fly adaptation to the exact equipment available and to user feedback mid-session.
- Built the platform on clean-architecture principles using a microservice setup on GKE (Kubernetes), with Swift on the client, LiveKit for low-latency voice, PydanticAI for agent adapters, React for web, and SQL plus Firestore for persistence.
- Took the product from zero to a few paying users. We didn’t figure out distribution before I left.
LLM Evaluation Voice AI LLM Conversational AI Python
Google
AI & Cloud Architect
July 2022 - November 2024 (2 years and 4 months)
Paris, France
Building AI and Cloud solutions for corporations launching new service offerings or looking to improve operational efficiency. Leading the development process from ideation to proof-of-concept and production deployment. Sample projects include AI agent development, AI-powered recommendations, data platform development and transition to cloud.
Google cloud Cloud Architecture AI Agents Voice AI Data Platform
MIT Media lab
Research Collaborator
ARCHITECTURE AND URBAN PLANNING
September 2020 - January 2022 (1 year and 4 months)
Cambridge, United States
Developed a bio-inspired framework for future mobility allowing autonomous vehicles to move more efficiently. Designed and built a vehicle-to-vehicle energy-sharing system, so future vehicles can share power while moving, extending their range in a decentralized and cost-efficient manner. Validated the framework through agent-based simulations.
Research Architecture and Urban Planning