I build infrastructure and orchestration for AI workloads, with 20+ years of experience delivering and operating software. Currently Principal Software Engineer at Midokura, in Barcelona. Earlier: observability inventory at New Relic, the internal platform at Adevinta, B2B features at Tuenti, and the music catalogue backend for Last.fm. If you work on AI infrastructure or software delivery, I would like to hear from you: anglorvaroa@gmail.com

Professional experience

(Since 2022): Principal Software Engineer at Midokura. We build systems to orchestrate, monitor and manage GPU servers and high-bandwidth network fabric for AI workloads, deployed in private datacenters. On an earlier project, I worked for ~3y on Sony's Edge AI sensing platform, managing fleets of cameras with intelligent sensors running AI vision models, and streaming real-time inferences to the cloud. Notable projects:

  • Design and implement systems to provision, manage and operate GPU infrastructure deployed in private datacenters, as well as multi-tenant IaaS/PaaS layers for AI workload orchestration.
  • Re-architected core backend systems (MQTT, Kafka, JVM-based) for real-time inference streaming. Achieved 100x performance improvements, with lower resource footprint.
  • Modernized release engineering and test strategy from fragile manual QA to automated delivery pipelines, reducing cycle time by 50%+ in complex embedded/cloud systems spanning distributed EU-JP teams

(2020-2022): Principal Software Engineer at New Relic. Embedded in internal platform teams managing entity inventory (hosts, containers, services, etc.), metadata and relationships.

  • Maintained critical infrastructure supporting most of New Relic's observability experience. Regularly in the critical path for ~50-75% of top-level company OKRs, and user-facing 99.8% SLAs.
  • Led design and implementation of high-throughput telemetry analysis system that processed millions of datapoints per second, inferring entity types and relationships from open, configurable rules.

(2016-2020): Staff/Principal engineer at Adevinta. Joined as lead in the Engineering Productivity team, and later became Technical Product Manager of Adevinta's internal developer platform. Served 1500+ engineers across major online marketplaces in Europe and South America.

  • Built developer tooling for the internal platform, including automated dependency propagation and a developer metrics platform. It grew organically, from 0 to hundreds of users with (measurable productivity gains).
  • Consolidated fragmented Infrastructure-as-a-Service projects (observability, kubernetes runtimes, etc.) into a cohesive PaaS that covered development, delivery and operations. (I published a piece about the platform).
  • Led the migration of Adevinta's online marketplaces from on-premise infrastructure and EC2 to Kubernetes-based PaaS, reducing runtime and operational overhead. Engineers from the local infra teams joined the central platform team, so their work served every marketplace, not only their own. (I published some technical articles about the experience.)

(2013-2016): Senior Software Engineer at Midokura. Core contributor and technical lead on distributed network virtualization system for OpenStack, implementing SDN control plane and data plane components.

  • Designed and implemented distributed network protocols in the network agent: L2/VxLAN gateways, NAT, Floating IP, Spanning Tree Protocol.
  • Integrated the control plane with hardware VTEP switches (feature docs).

(Prior to 2013): I worked as a software engineer building infrastructure for telcos, and backend services for high volume sites. At Last.fm (2007 to 2009), I managed the systems that ingested and served the largest streaming music catalogue in the world at the time, with ~60M tracks and ~40M users. At Tuenti I worked on backend systems behind groups, business and location sites.