AxS / Career record
Berlin, Germany / 2026

Principal AI/ML Platform Engineer

I make AI systems operable.

GPU Kubernetes, model serving, and internal tooling built for the engineers and agents that run production.

14 years

Across software, cloud, SRE, and AI platforms

$1.5M / yr

Observability cost removed

3,000+ services

Observed in production

300+ environments

Operated across AWS and GCP

01 / Selected impact

What changed after the platform shipped.

Architecture matters when it changes cost, autonomy, reliability, or the speed at which other engineers can move.

  1. 02AI platform

    Made AI a first-class platform workload.

    Built and operated Spryker's internal AI platform for multi-GPU, multi-LLM, and multi-agent workloads on EKS.

    One operating model for applications and AI.EKS / Karpenter / NVIDIA / LangGraph
  2. 03Observability

    Rebuilt the telemetry layer, then turned it into a product.

    Moved more than 3,000 distributed services from New Relic to OpenTelemetry, Prometheus, and the Grafana stack.

    $1.5M saved every year. Sold to five enterprise customers.OTel / Prometheus / Loki / Tempo / Mimir
  3. 04Agent operations

    Built tools that agents can operate safely.

    Developed multi-tenant MCP servers that let agents operate AWS, GCP, and Azure accounts, with end-to-end traces and metrics for every action.

    Agent-facing infrastructure with human-grade controls.Python / FastMCP / OpenTelemetry / Grafana Cloud
  4. 05Developer platforms

    Measured the platform by adoption, not feature count.

    Established self-service provisioning and golden paths at Spryker, AUTO1, and ING using Terraform, Backstage, and Argo CD.

    Hundreds of service teams adopted the paved roads.Terraform / Backstage / Argo CD

02 / Experience

From cloud foundations to production AI.

Each chapter expanded the operating surface: software, cloud, reliability, developer platforms, then model serving and agent infrastructure.

2012 - present
  1. Roam AI

    Principal Engineer, ML, Data and Platform Engineering

    Own the ML and data platform end to end: cloud infrastructure, backend services, data contracts, and model serving on hybrid GPU Kubernetes.

    Production Python and TypeScript / design reviews / mentoring and hiring

  2. Shipmoor

    Founder

    Building a code-integrity platform for agent-assisted engineering. Deterministic probes issue blocking verdicts; LLM inference never does.

    CLI / VS Code extension / agent harness / in-toto attestations

  3. Drizzle Systems

    Principal Consultant, Platform Engineering and MLOps

    Designed Roam AI's MLOps reference architecture and built Blocks Cloud's AWS platform, production MCP servers, and agent observability stack.

    Hybrid GPU Kubernetes / AWS platform engineering / agent infrastructure

  4. Spryker

    Senior Staff Engineer and Deputy Head of Cloud and SRE

    Designed the internal AI platform, led the observability rebuild, and operated 300+ customer environments and 3,000+ distributed services.

    Promoted from Lead Engineer after six months

  5. AUTO1 Group

    Team Lead, Site Reliability Engineering

    Led reliability for more than 5,000 microservices and shipped the SLO operating model and tooling adopted by hundreds of teams.

    Multi-account AWS / SLOs / reliability tooling

  6. ING / Lendico

    DevOps and SRE Lead

    Led the build and run of a regulated lending platform on Azure AKS, including custom Go operators, admission control, Vault, and observability.

    Promoted from Senior DevOps Engineer after six months

  7. Deutsche Bank / Yunar

    Cloud Platform and Site Reliability Expert

    Operated an Istio-backed Azure AKS platform of roughly 150 microservices and defined its SLI, SLO, and SLM-based platform offer.

    AKS / Istio / reliability architecture

  8. Smile Open Source Solutions

    Cloud and DevOps Consulting Engineer

    Delivered cloud engagements for Renault, Monoprix, and Svensk e-identitet across Azure, AWS, GCP, and OpenStack.

    Consulting / multi-cloud / open source

  9. Rosafi Holding + Tunisian Cloud

    Senior Software Engineer and Cloud Platform Engineer

    Built storage, database, and big-data platforms as self-service products, beginning a career at the seam between software and infrastructure.

    Python / distributed storage / data platforms

03 / Operating range

The tools follow the system.

Deepest in platform engineering and production operations, with enough software depth to build the control plane rather than just configure it.

  1. 01

    ML platform + serving

    KServe, vLLM, KubeRay, Kubeflow, MLflow, NVIDIA device plugin, GPU scheduling, LangGraph, MCP

  2. 02

    Kubernetes + cloud

    EKS, GKE, AKS, K3S, bare metal, Karpenter, KEDA, Istio, Linkerd, AWS, GCP, Azure, Hetzner

  3. 03

    Automation + delivery

    Terraform, Terragrunt, Pulumi, Ansible, Helm, Argo CD, Atlantis, GitHub Actions, GitLab CI, Backstage

  4. 04

    Observability

    OpenTelemetry, Prometheus, Grafana, Loki, Tempo, Mimir, Datadog, distributed tracing, SLOs, error budgets

  5. 05

    Languages

    Python, FastAPI, Flask, Celery, Go, Kubebuilder, TypeScript, Bash, SQL

  6. 06

    Data + orchestration

    PostgreSQL, Redis, Kafka, Airflow, Spark, Ceph, S3-compatible storage

04 / Beyond delivery

Teaching, disclosure, and foundations.

Sharing the operating lessons, and treating security as part of the platform.

Berlin / 2025 / Talk

Self-hosting DeepSeek-R1 on AWS EKS

A practical walkthrough of vLLM inference, KubeRay distributed scaling, and Terraform with Karpenter for GPU infrastructure.

MSRC / 2020 / Security

Microsoft Security Response Center acknowledgment

Co-disclosed an Azure Kubernetes Service privilege-escalation misconfiguration.

Education

2010 - 2012

M.Eng., Computer Systems Networking and TelecommunicationsENET'Com Sfax, Tunisia

2006 - 2010

B.Eng., Mathematics and Computer SciencePreparatory Institute for Engineering Studies, Monastir

Languages

FRFrenchNative

ENEnglishC2

DEGermanB1

Berlin / Open to a good systems conversation

Let's compare notes on production AI.