I build Cloud Infrastructure and developer tools with AI, in public.
Ten years shipping production software. Now building at the AI layer: MCP servers, agent workflows, and the small tools that make AI tools actually useful to engineers. Open source. In production.
Live projects.
All shipping, all weekly.
The current work, with status. Each one started as a real problem, got built in public on the channel, and is in use today.
An MCP server that connects Content Board (my Firebase PWA for YouTube production) to Cowork, Claude's desktop agent. I built it live on the channel, episode by episode, so the whole thing is documented end to end. It's the working proof that the MCP tutorials on this site come from a server I actually ship and use every week.
→ 02Chrome extension that adds a personal prompt library sidebar to ChatGPT and Claude. Save prompts with tone and format settings, insert them in one click, and sync across devices via Google Drive. No backend server.
→ 03A folder-based workflow automation system for Claude Cowork. Define repeatable AI workflows once: manual pipelines with approval gates, scheduled tasks, multi-stage projects. Includes a YouTube content pipeline and daily community scout out of the box.
→ 04A plugin for Claude Code and Cowork that prevents accidental file deletion and overwrites. Adds automatic backups, dry-run planning, and activity logging. A safety layer for AI-assisted development.
→The repeatable parts of working with AI tools.
When something works twice, I turn it into a checklist other engineers can run. The frameworks below are the ones I actually use, not aspirational.
5-Check
A pre-merge check for AI-generated code. Trust, verify, override before merge. Five gates between "agent finished" and "this lands on main".
AI-Assisted Development Loop
The outer loop the 5-Check sits inside. Task boundaries, review commands, security checks, manual testing, changeset review, human judgment: the whole arc, not just the merge gate.
Working with AI tools, on camera.
Short, technical, Hinglish. The experiments themselves, not summaries. The most useful ones usually include the part where it broke.
OpenClaw Tutorial: Build Your Own AI Agent (Setup to Security Explained)
How OpenClaw actually works internally, why it burns 10,000 tokens before your first message, and what makes it fundamentally different from Claude Code or Cowork. Real use cases, honest security risks, and whether you should try it today.
Watch
I Built an AI Website Monitor in Minutes: Claude Cowork Tutorial (No Code)
How Claude Cowork automates the entire workflow (reading websites, analyzing content, and delivering summaries directly to Slack) without a single line of code. The Brain, Eyes, Mouth mental model for AI automation.
Watch
Git Worktree + Claude Code: Run 3 AI Agents on One Repo
One terminal, one agent, 15-minute wait per task. What if you ran 3 AI agents in parallel on the same codebase? Production-grade workflow for 3-4x development speed without sacrificing code quality or safety.
WatchProduction work before the AI layer.
The boring depth. Identity migrations, multi-tenant Kubernetes, legacy modernization, probabilistic deployment models. That's the systems work behind knowing when AI helps and when it doesn't.
AI-Powered Release Orchestration for Multi-Tenant SaaS
Probabilistic models making real deployment decisions. AI in production before the hype cycle. $17K saved annually.
Case → 2024 — PresentZero-Downtime Migration to AKS for 400+ Tenants
Multi-tenant Kubernetes migration, version lock-in solved. Today I'd automate the tenant-by-tenant validation with AI. Back then we did it manually.
Case → 2021 — 2024How We Rebuilt a University’s Legacy Credit System with Cloud-Native Architecture – and Boosted Revenue by 30%
Cloud-native re-architecture of a payments system. 30% revenue lift. The kind of domain complexity where AI-generated code breaks on day one if you don't understand the system.
Case → 2024 — PresentHow We Future-Proofed User Management with a Scalable Azure Migration
Identity systems have zero margin for error. AI accelerates the boring parts; a human owns the critical path.
Case →If you're putting on a serious AI-builders event, I'm in.
Practitioner talks, technical workshops, deep dives on MCP / agent architecture / AI-assisted dev workflows. No keynote fluff. I'd rather sit on a panel, run a hands-on, or walk through a real build.
Want a summary in your model of choice?
Drop the brief into Claude, ChatGPT, Gemini, or Perplexity. Each link opens a fresh chat with a research prompt about my work pre-loaded.