Frédéric HAMPEL - Applied AI Engineer

Summary

Applied AI engineer and DevOps lead driving LLM initiatives end to end. Early adopter who pushed AI into production workflows and ticketing, handling infra, CI congestion and security issues, and implementing the first internal LLM features. Most of my code is now written by coding agents that I review, which lets urgent production fixes ship in about 30 minutes.

Experience

Oct 2023 - present

Cycloid · Senior DevOps Engineer

French Tech company selling a platform engineering and internal developer portal product as SaaS, fully remote with an English-speaking internal team and international (incl. US) customers. Joined as DevOps; grew scope to the CLI, the Terraform provider, the SaaS and on-prem platform, and technical lead on large enterprise accounts.

Owned solution engineering for two of the company's largest enterprise accounts, designing and implementing the bridge between customers' GitLab projects and pipelines and Cycloid's Concourse CI so GitLab features (review apps, environments, managed deployments) worked end to end; drove the product developments these customers needed, and led the integration for Cycloid's largest contract to date
Drove AI-assisted development across API, CLI, Terraform provider and stacks: about 1,250 commits merged in 13 repositories in 2026, about 70% written by coding agents
LLM ticket-type classification in the customer helpdesk, evaluated on real production tickets: 84.5% accuracy against 55% for keyword rules, with the cheaper of two models shipped at nearly the same quality. PII redaction before any model call, rate limits and a keyword fallback; GDPR analysis
Company LLM gateway (Bifrost on Kubernetes) for 3 internal apps, with per-app budgets, rate limits and provider fallback. Replaced an end-of-life model and migrated 6 production keys with no downtime
Helpdesk MCP server: 60 tools exposing the whole support portal to AI agents over streamable HTTP, with per-user revocable API keys and a matching CLI
Maintained multi-tenant SaaS and on-prem platforms on Kubernetes and VMs, automated infrastructure with Terraform, Concourse and Cycloid CI, and cut cloud spend with per-developer ephemeral environments

Feb 2023 - Oct 2023

WedoLow · DevOps Engineer and Head of IT

Start-up of about 10 people building embedded software optimisation tools, Rennes.

Packaged the product (Flutter desktop app and on-prem server) with Windows and macOS installers, and a Go CLI that cut an on-prem deployment from about 8 manual steps to 3
Company IT and security questionnaires for large automotive and telecom customers

Jan 2021 - Feb 2023

Creative Group · DevOps Consultant

IT services company, Rennes. Placed on customer missions.

Leocare (insurance scale-up, 10 months): moved production from scattered VMs to a hardened Azure platform built with Terraform and Ansible; cleared the new CISO's security backlog, added high availability and zero-downtime deployments, and set up monitoring and central logging (Grafana, Loki)
IDnow (AriadNext, identity verification, 10 months): production releases and on-call on an on-premise Debian server fleet, plus automated system patching, DNS updates and certificate rotation
Nehs Digital (3 weeks): Ansible playbooks applying ANSSI security recommendations to about 800 machines

Oct 2019 - Oct 2020

Maincare Solutions · Work-study

Medical software company building national-scale electronic medical record systems. Work-study role focused on PostgreSQL, high availability and automation.

Designed and built a PostgreSQL high-availability cluster with Patroni, ZooKeeper and HAProxy for a country-wide medical records customer, focused on consistency and failover so the service stayed available through node failures
Upgraded PostgreSQL from 9.4 to 11, validated with load tests, and turned the whole lifecycle (install, provisioning, migration, decommissioning) into infrastructure as code with Ansible instead of ad-hoc scripts
Wrote a Python tool to manage SQL backup files directly on customer machines, making backups more reliable and repeatable for on-prem hospital deployments

Personal projects

Selected AI projects

Software factory running 24/7: small open-weight models write the code, frontier models plan and review. Agents are scheduled around local inference limits on how many can run at once
Auto mode for coding agents: static rules first, then a small local classifier as a fail-closed last resort; 0 false allows on 150 cases
Dionysos: a live-performance DAW in Rust for my band's rehearsals and shows (real-time JACK audio, click tracks, mixer buses, backing tracks), built with coding agents

Personal infrastructure

Kubernetes cluster at home running about 50 applications, all as code (Nix, Terraform), Renovate updates, GitLab CI deploys with automatic rollback
Secrets and SSO: OpenBao with External Secrets Operator, single sign-on across all tools
Observability and backups: Prometheus, Grafana, Loki, Tempo; encrypted off-site backups with restic
Self-hosted LLM gateway with per-client access control and privacy tiers deciding which providers a client can reach

Skills

LLM systems: LLM gateway (Bifrost), model routing and provider fallback, LLM evaluation against baselines, structured output validation, model selection by cost and quality, budgets and rate limits
Agents and tooling: MCP servers (design and implementation), coding agents (Claude Code, Codex, opencode, pi), agent harnesses and skills, multi-agent orchestration, human-in-the-loop approval flows
Inference: local inference (llama.cpp, Vulkan), open-weight models (Qwen, DeepSeek), quantization, latency and throughput benchmarking
Safety and governance: PII redaction before model calls, GDPR for LLM features, model allowlists, permission classifier for agent tool calls
Platform: Kubernetes, Helm, Terraform, NixOS, GitLab CI, Argo CD, OpenBao and External Secrets Operator, Prometheus and Alertmanager, Grafana, Loki, AWS, Azure, GCP, Scaleway
Languages: TypeScript, Python, Go, Nix, Bash

Languages

French (native), English (fluent), German (basic)

Education

2017 - 2019

Bac+2, IT maintenance and support (Gestionnaire en maintenance et support informatique), CESI Dijon

Work-study at SNCF (French national railway company), focused on infrastructure and virtualisation.

SNCF legacy tooling: virtualised maintenance workstations on Linux with KVM to run systems from MS-DOS to Windows 7, forwarding serial ports so technicians could connect to locomotives without relying on ageing physical machines
“Projet Évolution”: designed the infrastructure migration of a company growing from 80 to 120 employees and moving to new offices, proposing a Proxmox server cluster and thin clients for most workstations

2013 - 2016

Sound and lighting technician diploma, Music Academy International, Nancy

Outside work

President of our school's parents' association (since Nov 2026) · Singer and bass player in a live band