Cloud · DevOps · Enterprise AI Infrastructure
Cloud platforms, Kubernetes clusters, and private AI systems that hold up in production.
Royal Technology is the engineering practice of a cloud and infrastructure engineer with more than a decade in enterprise IT. I move organizations to Azure without losing data, run Kubernetes the way production demands, and get large language models serving fast on hardware you own.
Capabilities
Three practice areas, one standard: changes go through version control, everything is measured before and after, and nothing ships untested.
Cloud platforms and DevOps
Design, migration, and operation of cloud environments, with the automation to keep them reproducible.
- Azure IaaS/PaaS architecture and organization-wide migrations, from planning through cutover
- Terraform and OpenTofu modules for multi-tenant infrastructure under version control
- Azure DevOps and GitHub Actions pipelines for hands-off releases
- Entra ID and Microsoft 365: identity, conditional access, and Intune at fleet scale
- Azure FinOps reviews that find waste; one audit cut a client's footprint by 75%+
- Azure Virtual Desktop with FSLogix profiles and hosts that scale to zero after hours
Kubernetes and platform engineering
A Talos Linux cluster runs around the clock as the proving ground for everything that ships to clients.
- Immutable Talos Linux nodes, Flux GitOps, and Flagger canary rollouts
- Istio service mesh with mTLS over Cilium, zero anonymous routes
- LINSTOR and ZFS distributed storage; CloudNativePG with failover and WAL archiving
- Velero backups to S3, External Secrets Operator, and Bitwarden Secrets Manager
- Kyverno policies, Falco runtime detection, and Trivy image scanning
- LGTM observability: Loki, Grafana, Tempo, and Mimir with alert routing
Local AI inference and optimization
Enterprise-grade language model serving on your own hardware, so sensitive data never leaves the building.
- Models quantized to W4A16 AWQ and served with SGLang
- Tensor parallelism across NVLink-paired GPUs at 96k-token contexts
- MTP/NEXTN speculative decoding and context-pool tuning worth +30% single-stream throughput
- Benchmarks before and after every change; long-context configs probed past 90% load before shipping
- GPU power capping to 250W per card: less heat, minimal throughput cost
- LiteLLM gateway with per-user keys, spending caps, and Langfuse token-level tracing
- Speech-to-speech stacks on pipecat with Silero VAD and on-prem ASR
- Guardrails on every request path, including NeMo Guardrails and model-level checks
- Fine-tuning on distilled, curated datasets; adapters gated on holdout tests written first
Selected work
Clients have ranged across healthcare, government, financial services, real estate and title work, and industrial distribution. Who they are, and how their environments are built, stays confidential; several operate under regulatory and contractual obligations that require it.
Cloud migrations without data loss
15+ organizations moved to Azure IaaS/PaaS with 500TB+ of data delivered intact. The largest single migration moved 30TB, and every runbook stayed in the client team's hands.
Virtual desktop platforms
VDI pools with role-based least-privilege access, roaming user profiles on managed file storage, and session hosts that scale to zero when idle. Compute cost drops to nothing outside working hours.
Hybrid data integration
Self-hosted integration runtimes feeding on-prem operational databases into cloud analytics. Every pipeline definition lives under version control, so connection changes are reviewable.
Endpoint management at fleet scale
Policy design and onboarding automation across tens of thousands of managed endpoints on macOS, Windows, and Linux. A prior role covered 4,200+ endpoints for 100+ clients, so fleet scale is familiar ground.
Identity isolation at 100+ tenant scale
Tenant separation, scoped administrative units that cap blast radius, and hybrid identity sync with cloud-based single sign-on across multi-organization estates.
Security and cost audits
Exposed remote-management paths closed down, environments redesigned for least privilege, and one engagement that cut cloud footprint by 75%+ plus thousands from the monthly bill.
How engagements work
Scope is agreed up front, work lands in your repos, and you keep the runbooks.
Fractional platform engineering
Architecture, deployment, and operations for Azure estates and Kubernetes clusters, owned end to end.
Migration delivery
Cloud, identity, and email migrations, with documentation your team can keep and reuse.
AI infrastructure
Assessment, buildout, and tuning of on-premises LLM serving, from GPU selection through serving benchmarks.
Every engagement ships with documentation. If that sounds like the gap you're trying to close, get in touch.
Contact
Tell me what you're working on and I'll get back to you by email.