Bridging dev and ops with code
For three and a half years I have owned AWS, Kubernetes, CI/CD, observability, on-call and security
at one startup through three product generations: the MetaOne Web3 wallet, the Saakuru blockchain
and a multi-LLM AI agent platform. Eight years across software and infrastructure in total.
The results I point to. I cut recurring AWS spend by about 40% with no-upfront reservations,
Fargate Spot, S3 gateway endpoints and the removal of resources nothing used. I built a daily
stability report for 7 production apps that ran 157 times unattended over five months, delivered
95.5% of the time, and raised its own alert on all 7 failures. I found and fixed a two-month silent
data-loss bug in production error tracking after ruling out three other causes. I ship
infrastructure changes through a researcher, worker and tester loop of Claude Code agents, and the
tester has to show a failing run before I accept a fix.
Today the live estate runs on Vercel, Fly.io, Supabase and Cloudflare with self-hosted
Grafana, Loki and Prometheus, plus three AWS accounts under cost monitoring. I define the SLOs,
SLIs and error budgets, carry the pager, write the postmortems and keep the runbooks. When an LLM
or agent misbehaves in production, the debugging is mine, across
Anthropic Claude, OpenRouter, DeepInfra and Perplexity.
Before infrastructure I spent four and a half years building backend and frontend systems in Laravel,
Node.js, Angular and Vue, which is why I can debug the application as well as the machine under it.
Where I've worked
One employer and one continuous contract with AAG Ventures. The business moved from a Web3 wallet to its own blockchain to AI infrastructure, and I owned DevOps through every stage. The three periods below are the same job.
- Escalation point for production incidents, including LLM and agent incidents, on a platform integrated with Anthropic Claude, OpenRouter, DeepInfra and Perplexity: about 30 agent deployments across production and test, each with isolated ECS compute and its own Supabase database.
- Built and ran two K3s clusters (Singapore and EU) after the engineer who started them left, and operated a third: 33 workloads under Flux CD with Helm and Kustomize, SOPS-encrypted secrets, Longhorn storage backed up to S3, Traefik ingress, cert-manager, Prometheus and Loki. Handled etcd disaster recovery, certificate rotation and OOM incidents on live clusters. (Oct 2025 – May 2026)
- Daily stability report for 7 production apps: a GitHub Actions job that pulls Vercel deploy state, GlitchTip errors and Grafana metrics and posts one Slack report per app each morning, with a dead-man's switch. 157 unattended runs over five months, 95.5% delivered, all 7 failures raised their own alert.
- Agents in the delivery loop. Infrastructure changes go through a researcher, worker and tester loop of Claude Code agents that hand off through a task document; the tester has to show a failing run before a fix is accepted. The loop caught a report that built 103 Slack blocks against a limit of 50 before it reached production.
- Found and fixed a two-month silent data-loss bug in production error tracking, where every event was accepted, counted and discarded, after ruling out three other causes.
- Evaluated AWS Bedrock and ruled it out to keep the platform portable across providers, then led the move from AWS to Vercel, Fly.io and Supabase, leaving one client production account and two archives on AWS under cost monitoring.
- Observability and on-call: Prometheus, Loki and Grafana over 45 scrape targets; SLOs, SLIs and error budgets; postmortems and 11 runbooks.
- AWS as code and CI/CD: 22 Terraform modules across four repositories; 197 GitHub Actions workflows across 65 repositories, including one service deploying four environments across two AWS regions.
- Security: application secrets into HashiCorp Vault, production SSH behind Teleport, third-party IAM scoped to bucket ARNs, MFA audit of 17 Supabase projects.
- Managed high-availability blockchain RPC endpoints, nodes and Blockscout explorers for the Saakuru chain and wallet (100K+ installs, 1M+ transactions), through two chain upgrades with services kept available.
- Cut recurring AWS spend by about 40% after the 2023 Savings Plans expired unrenewed: no-upfront reservations for the steady EC2 fleet, Fargate Spot for tolerant ECS services, S3 gateway endpoints to keep bucket traffic off the NAT gateway, and removal of resources nothing used.
- Diagnosed distributed system failures: chain reorgs, transaction inconsistencies, indexing delays and explorer mismatches.
- Supported third-party node applications on Azure VMs alongside the primary AWS estate, including monitoring and operational maintenance.
- Helped decommission legacy blockchain stacks during the company-wide pivot to AI-focused infrastructure.
- Automated multi-environment deployments using Jenkins and GitHub Actions, reducing manual overhead and deployment errors for the MetaOne Wallet.
- Stabilized production environments by performing deep-dive troubleshooting into Node.js services and PostgreSQL database performance.
- Gained cross-cloud proficiency by managing AWS EC2/S3 assets and supporting initial Azure-based infrastructure operations.
- Standardized dev-to-prod parity by containerizing Laravel and Vue.js applications using Docker, improving developer onboarding speed.
- Developed internal REST APIs and admin dashboards, providing a strong software engineering foundation for future infrastructure automation roles.
Technical toolkit
A pragmatic toolkit built across cloud, container orchestration, IaC, and observability.
What I've built and run
Four things that say more about how I work than a tools list does.
Education & certifications
Certifications
Currently working toward cloud and Kubernetes credentials. This list grows.
Open for side gigs
Available for short-term DevOps consulting, infrastructure audits, and freelance contracts — on top of my full-time role. Outside of business hours & weekends.