I Built a Personal AI System That Runs My DevOps Career

Six months ago, every time I started a new Claude Code session, I had to re-explain everything. My clusters. My apps. My playbooks. My preferences. Every. Single. Time.

That friction wasn’t just annoying — it was costing me hours every week. As a Platform Engineer managing 14 apps across multiple OpenShift clusters, the context I carry in my head is enormous. And AI assistants, for all their power, have the memory of a goldfish.

So I built Cortex.

What is Cortex?

Cortex is a persistent, multi-machine AI memory and automation system built entirely in plain markdown. It’s my “second brain” — but instead of being a passive note-taking app, it actively powers my AI-assisted workflow.

Here’s what it does:

  • Remembers everything: cluster topologies, app architectures, shared service dependencies, Helm chart patterns, secret rotation schedules
  • Runs playbooks: safe, repeatable procedures for common ops tasks (pod restarts, Helm upgrades, secret rotations, morning briefings)
  • Syncs across machines: one cortex sync command pulls the latest knowledge to any machine
  • Survives tool changes: it’s just markdown files in a git repo — if Claude disappeared tomorrow, my knowledge stays

The Architecture

cortex/
├── personal/         # portable across jobs
│   ├── reference/    # helm patterns, k8s fundamentals
│   ├── life/         # career, finance, health, learning
│   └── projects/     # homelab, brand building
│
└── clients/          # strictly isolated per employer
    └── bcgov/
        └── memory/   # 44 files, 3000+ lines of operational knowledge

The key insight: strict client/personal separation. My Helm patterns and career goals follow me everywhere. My employer’s cluster details never leave their boundary. This makes the system safe to use long-term across multiple jobs.

Real Impact

Before Cortex, a typical morning looked like this:

  1. Open terminal
  2. Try to remember which tickets I was working on
  3. Re-explain my cluster setup to Claude
  4. Figure out which environments need attention
  5. Slowly get into flow

After Cortex:

  1. Open terminal
  2. cortex sync
  3. “Run morning brief”
  4. Claude reads my state files, checks Jira via MCP, scans clusters
  5. I get a prioritized action list in 30 seconds

That’s a 50% reduction in daily startup friction. Multiply that across weeks and months, and it compounds massively.

The Playbook System

The most valuable part of Cortex isn’t the memory — it’s the playbooks. Each playbook encodes a safe, repeatable procedure:

  • Helm Upgrade Playbook: template first, dry-run second, backup third, execute fourth. Never skip a step.
  • Secret Rotation Playbook: extract → modify → re-encode → patch → verify. One wrong move could take down production.
  • Morning Brief Playbook: pull Jira tickets, scan cluster health, check state files, prioritize the day.

These aren’t just documentation. They’re executable instructions that Claude follows step-by-step, with safety gates at each stage.

Why Plain Markdown?

I’ve seen too many “second brain” tools come and go. Notion, Obsidian, Roam — they’re all great until the company pivots, raises prices, or shuts down.

Cortex is built on three things that will never disappear:

  1. Git — version control that’s been stable for 20 years
  2. Markdown — readable by humans and every AI model
  3. The filesystem — the most universal interface

No vendor lock-in. No proprietary formats. No subscriptions. Just files.

The Bootstrap Problem

The hardest part wasn’t building Cortex — it was making it frictionless. If syncing your memory requires 5 manual steps, you’ll stop doing it within a week.

The solution: a single cortex sync command that:

  1. Pulls the latest from git
  2. Re-runs the bootstrap
  3. Copies memory files to the right locations
  4. Renders config templates with machine-specific paths
  5. Configures Claude Desktop with MCP servers

One command. Zero thinking. That’s the bar.

What’s Next

Cortex is one week old and already handling my daily DevOps workflow. The roadmap:

  • Month 1: All five core playbooks battle-tested on real work
  • Month 2: Gmail and Calendar integration (morning brief pulls calendar too)
  • Month 3: Content drafting pipeline (this blog post was drafted with Cortex)
  • Year 1: Full personal brand automation — blog, newsletter, social media
  • Year 2: Mobile access, voice interaction, always-on assistant

The substrate is built. Now it compounds.

Try It Yourself

You don’t need to build something as elaborate as Cortex. Start with one thing:

Create a MEMORY.md file in your project directory. List the 10 things you re-explain to your AI assistant every session. That alone will save you hours.

Then build from there. Add playbooks. Add state files. Add a sync command. Let it grow organically.

The best system is the one you actually use. Start small, compound daily.

Rohith Yadla is a Platform Engineer building at the intersection of infrastructure and AI. Follow the journey at yarova.ca.