The Lab

Specialist infrastructure, in public.

The deep work behind Sirius DevOps: GPU inference at scale, GitOps-driven clusters, and a homelab that runs production. This is the proof page for hiring managers, specialist consulting, and AI-infra buyers.

AI Infrastructure

AgentForge — vLLM on Kubernetes.

A cloud-native platform for high-throughput vLLM inference across dual-tier NVIDIA GPUs. 100% Git-driven via Terraform + Flux CD, with KEDA autoscaling that cut GPU compute spend ~70%.

AI Infrastructure

vLLM serving, KEDA scale-to-zero

  • vLLM serving Qwen models up to 27B on L4 / RTX 6000 / A100
  • KEDA scale-to-zero — idle reasoning capacity cost nothing
  • Zero-touch GitOps: every change audited and synced from Git
  • Per-tenant isolated databases; automated security audit-to-fix
  • 1,000+ commits, fully public on GitHub — github.com/sirius0xdev/lab
Blockchain Infrastructure

Multi-cloud Kubernetes — Zcash node stack.

A cluster bootstrapped from bare nodes with kubeadm — no managed control plane — running a full Zcash stack as first-class workloads.

Kubernetes / GitOps

kubeadm across clouds, Flux-managed

  • kubeadm bootstrap across multiple cloud providers
  • Zebrad consensus node + lightwalletd gRPC service
  • Cilium BGP control plane + L2 announcements, single LB IP pool
  • Flux CD GitOps with SOPS/age secret decryption
  • CloudNative PG 3-node HA cluster underneath it all
  • Full manifests public → github.com/sirius0xdev/lab
Raw Screens

Actual sessions. Not mockups.

Unretouched terminal captures from real operations on my clusters. Cluster identifiers are redacted; everything else is exactly as it happened.

k9s tailing HTTP logs from an AI agent service running in-cluster while its web UI responds in the browser
agent ops AI agent served from Kubernetes. Web UI on the left; k9s on the right streaming the same service's request log — every call returning 200 from inside the cluster.
k9s showing a model-puller job downloading a 42GB model at 982 MB/s
model delivery 42 GB model pull at ~982 MB/s. In-cluster init job streaming a DeepSeek analyst model onto GPU nodes — watched live through k9s.
k9s tailing vLLM inference server logs with throughput and KV-cache metrics
inference Live vLLM serving Qwen3. Real-time inference metrics: generation throughput, GPU KV cache usage, prefix cache hit rate — plus tool-call parsing on a multi-tenant deployment.
second k9s session tailing vLLM API traffic with 200 OK responses
inference API traffic under load. POST /v1/chat/completions requests completing 200 OK while speculative-decoding acceptance rates stream past — multi-tenant namespaces isolated per customer.
Real-Time Systems

Trading platform & quant pipeline.

Event-Driven Systems

Trading Platform

Event-driven microservices on GKE — Next.js dashboard, Hyperliquid + Solana execution, Kafka streaming, Redis caching, PostgreSQL 17. mTLS, cert-manager, default-deny network policies, full CI/CD.

Autonomous Pipelines

News Quant Pipeline

Scraped 371 global financial feeds, ran DeepSeek-R1 analysis locally, and generated trading signals — all VPC-private, all self-running via CronJobs.

Self-Hosted Edge

24/7 Bare-Metal Homelab

Raspberry Pi edge server running continuously: Forgejo git + self-hosted CI/CD runners, nginx routing behind a Cloudflare tunnel, containerized client sites, and a Telegram-relayed contact API. The same GitOps push-to-deploy pipeline that serves this page.

Stack

The tools behind the work.

Infrastructure

Terraform1.7+
GKE / AWSCloud
DockerContainers

Orchestration

Flux CDGitOps
CiliumeBPF
Helm / KustomizeDeploy

AI / ML Infra

vLLMServing
NVIDIA GPUsL4 · A100 · 6000
KEDAScale-to-zero

Database

PostgreSQL15+
Redis7
KafkaStreaming

Security

SOPS + AgeSecrets
mTLSCerts
Cilium NetPolDefault-deny

Observability

PrometheusMetrics
GrafanaDashboards
k9sCLI
Next step

Need this for a product team?

Book a review and we'll talk about what your stack actually needs.

← Back to services