Story / Why the experience matters
From datacenters to agents.
More than fifteen years designing, building and operating what runs underneath—from physical systems and private clouds to platforms serving people across more than one hundred countries.
Understanding infrastructure from the bottom up.
The progression began with systems administration and moved into the physical and network layers of a Tier III datacenter.
- Supported the first Lumière Festival as a system administrator.
- Founding member of DCforDATA, later acquired by Free (Iliad).
- Worked across power, cooling, availability, security, Layer 2/3, RIPE, BGP, DNS and customer operations.
Infrastructure becomes software.
Private cloud and scientific computing made infrastructure programmable, measurable and repeatable.
- Built and operated OpenStack and HPC environments at CNRS for eTRIKS and CERN-related work.
- Worked on distributed systems, performance optimization and an R computing farm.
- Continued into GCP, Kubernetes, CI/CD, databases, monitoring and security at Optimiz.me.
Infrastructure becomes somebody else’s foundation.
The focus shifted from operating systems to building shared platforms that teams and operational processes could depend on.
- Led SRE work supporting logistics across 15 Vente Privée warehouses operating continuously.
- Worked on core storage, virtualization and on-premises Kubernetes for engineering teams.
- Built vehicle OS pipelines and fleet-management infrastructure at Navya, then cloud migrations and Kubernetes tooling through orchestrato.io.
Reliability while everything is growing.
At BeReal and TenTen, the engineering problem was shaped by rapid growth, short traffic windows, low latency and production reach.
- Built Google Cloud infrastructure through BeReal’s growth into tens of millions of daily active users.
- Addressed availability, latency, throughput, observability, security and cost-efficiency constraints.
- As a founding engineer and Head of Platform at TenTen, built AWS and low-latency WebRTC foundations used across more than 100 countries.
Infrastructure for autonomous systems.
PixelMochi applies the same reliability and platform disciplines to AI-native startups and to the control problem created by autonomous agents.
- Consulting focus: SRE, cloud, performance and reliability.
- Working style: agent-driven, specification-first, ADR-backed and multi-model.
- Product question: how agents execute while identity, policy, data and infrastructure boundaries stay explicit.
Technical breadth
Tools change. Engineering domains accumulate.
Infrastructure
Kubernetes · AWS · GCP · OpenStack · bare metal
Data
PostgreSQL · ScyllaDB · ClickHouse · Redis
Observability
Prometheus · Grafana · Datadog · ELK
Automation
Terraform · OpenTofu · Ansible · Pulumi
Engineering
Go · Python · TypeScript · Bash
AI / ML
Agent-driven workflows · vLLM · llama.cpp · CUDA
One experience, two directions
Apply the lessons. Encode the lessons.
Services
Apply infrastructure experience directly to a client system and team.
Product
Explore how those principles become infrastructure for agents.
Infrastructure is a progression, not a stack list.
If your current system sits somewhere along that progression, let’s discuss the constraint.