Arnault Bretagne

Data Engineer · Agentic AI

Profile

Generalist engineer turned Data Engineer, I combine an understanding of business needs, ELT pipeline development, financial reporting and team leadership. Outside work, I design and run Agora, a platform for sandboxed coding agents on my own Kubernetes cluster: from scoping to production, managing the agents as a team.

Professional experience

Data Engineer

FHVI · Lausanne · Aug. 2024 - Present

Maintenance and development of the operational data hub on SQL Server, in an on-premise environment.

  • LLM: Led the LLM proof of concept.
  • Ingestion: Architecture of the new ingestion solution.
  • Cloud: Feasibility study of a migration to the cloud.
  • Infrastructure: Administration of the backend and the virtual machines.

Data Engineer

Capgemini · Geneva · Nov. 2021 - Jul. 2024

Consultant / Senior Consultant

SAP financial reporting and planning for a European semiconductor leader.

  • Leadership: Business analysis, specifications, team of 5 developers.
  • ELT / SQL: Pipelines on SAP HANA Cloud and SQL optimisation.
  • Modelling: KPIs, cost allocation and valuation of the production backward schedule.
  • ML: Cost classification with SAP Data Intelligence and hana-ml.

Buyer

Losinger Marazzi · Lausanne · Feb. 2016 - Dec. 2020

Procurement of building services (HVAC, plumbing, electrical), kitchens and floor coverings; innovation.

Personal projects

Agora - Agent platform

agora

Design, development, operations · Jul. 2026 - Present

A web interface to work with coding agents that run in isolation, find all their work again after an interruption, and have them work as a team.

  • Executions: Claude Code, Codex and opencode over ACP, each agent in its own Kata microVM, warm pools.
  • Secrets: no credential in the sandbox; a gateway (agentgateway) adds them to each request, under permissions signed by Agora.
  • MCP: agents search and reread past work, and delegate to other agents, under human approval.
  • Teams: on Prefect, a lead splits and staffs the work, developers deliver pull requests, a separate judge verifies; I approve and merge.
  • Log: every exchange written to PostgreSQL before acting on it; recovery after a loss, full-text search.
  • Reliability: acceptance cases per spec, proven locally, on the cluster or live, with dated evidence.

bretagne.dev - Kubernetes

infra-k8s

Self-hosted GitOps platform · Apr. 2026 - Present

The cluster that runs Agora and my services on a mini PC at home: every change goes through a pull request, and Flux applies it.

  • GitOps: Flux, SOPS, Renovate; protected main branch: agents propose, I merge.
  • Security: passkey SSO (Pocket-ID), a Cilium network closed by default, Kata microVMs for untrusted code.
  • Previews: a label on a pull request creates an isolated environment, with a copy of production data when needed.
  • Data: PostgreSQL (CloudNativePG), off-site backups, a restore tested every night.
  • Observability: Prometheus, Loki, Grafana and alerts.

Certifications

  • DP-203 · Data Engineering on Microsoft AzureOct. 2021
  • dbt FundamentalsDec. 2023

Education

Arts et Métiers ParisTech - Aix-en-Provence / Paris

General engineering degree

  • Industrial management and global supply chain
  • Mechanical and industrial engineering