Fourteen years designing, automating and running cloud infrastructure — AWS, GCP and Azure, Kubernetes, Terraform — for teams who need it to stay up. Lately I've been building my own: three open-source products, shipped end to end and running in production today.
Three products taken from an empty repo to something people actually run — the infrastructure, the product surface and the operations. All open source: read the code before you read my CV.
A full cloud platform, built solo: multi-tenant compute with hardware isolation, managed Postgres with point-in-time restore, git-driven app hosting with blue-green deploys, serverless functions, two SDKs and usage billing — on GCP, with Terraform and Ansible underneath. Workloads boot in 179 milliseconds.
Prompt → a running full-stack app, then fork it. A self-hostable Lovable/v0 alternative where generated apps run in real microVMs with real databases — and you can branch a running app and its data to explore variations in parallel.
A drag-and-drop canvas for AI agent workflows. Wire LLMs, databases, APIs and code-execution nodes together and deploy — with code nodes running in isolated microVMs so untrusted code never touches the orchestrator.
Long enough to have seen most failure modes twice — and recent enough to have built a production platform from scratch this year.
Cloud architecture, Kubernetes, Terraform, CI/CD, observability and the kernel-level plumbing beneath it. I've owned every layer, so I know where they break each other.
Solo on-call for a multi-tenant platform. That shapes the designs I bring you: blast radius, recovery paths, and alerts that fire before customers notice.
A complete cloud platform in about six months, alone. Public commit history — you don't have to take my word for the pace.
I pay my own cloud invoice, so cost is an engineering constraint to me. Scale-to-zero and usage billing exist because idle infrastructure offended me.
Not a keyword list — this is what I operate in production right now.
Platform engineering, cloud architecture, and SRE work — for teams who need someone who has operated what they're about to build.
GCP, AWS and Azure architecture. GKE/EKS, autoscaling, multi-region, migrations — design review or hands-on build.
Terraform, Ansible and CI/CD that reproduces your infrastructure instead of describing it. Blue-green deploys and clean rollbacks.
Alerting that catches failures before customers do, plus the unglamorous work of cutting a cloud bill in half.
The rare specialty: hardware-isolated compute, snapshot/restore, egress policy. If your product executes code you didn't write, this is my day job.
Deep technical posts written from incidents I had to fix myself.
What a world model is in 2026 and what actually shipped: Genie 3 at 720p and 20 to 24 fps for a few minutes behind a $249.99 subscription, World Labs' Marble and Atlas and the $8.2 billion all-stock sale to AMD, NVIDIA Cosmos 3 at 64B, 16B and 4B under OpenMDW 1.1, V-JEPA 2 planning a robot action in 16 seconds against Cosmos's four minutes, LeCun's AMI Labs raising $1.03 billion at $3.5 billion, the Physics-IQ ceiling of 29.5 out of 100, what inference costs, and a V-JEPA 2 embedding script I ran on a laptop CPU.
14 minOct 5, 2026The no-code automation tools became agent runtimes this year and the security record followed: n8n went from a EUR 250M valuation to a reported $5.2B while publishing 196 GitHub security advisories in 2026 (19 critical), a CVSS 10.0 unauthenticated flaw left 59,558 hosts exposed per Shadowserver, and CISA added one n8n and six Langflow CVEs to its exploited list. Zapier MCP counts 25 million tool calls, Copilot Studio bills 5 credits per agent action, Agentforce ARR passed $1.5B, and Temporal raised $550M at $12.55B with 4,300 paying customers. Plus a hardened n8n queue-mode compose file.
15 minOct 5, 2026What the model-collapse research tested versus the headline: Shumailov's Nature loop deletes real data every generation and takes OPT-125m from 20 to 28 perplexity, while accumulating data bounds the error; my 100-generation Gaussian experiment shows median variance falling from 0.98 to 0.36 under replacement and holding at 0.98 under accumulation; how Phi-4 (40% synthetic), Llama 3, DeepSeek-R1 (800k samples), Qwen3, Kimi K2 and Nemotron-CC-v2 (38%) actually use it; Epoch's 300T-token stock, Graphite's 49.9% AI web, Cloudflare's 38,065:1 crawl ratio and the $1.5B Bartz settlement.
15 minI take a small number of consulting engagements, and I'm open to select full-time infrastructure roles — remote.
Platform engineering, cloud architecture, SRE. Project-based or monthly retainer.
$75–100/hr · $6–8K/mo retainer
What I deliver →Senior/staff/principal infrastructure, platform or SRE. Remote, with US or EU overlap.
Go · Kubernetes · Terraform · GCP/AWS/Azure
LinkedIn →Tell me what's slow, expensive or on fire. I reply to everything that isn't a template.
hello@ajayk.sh
Email me →