From command not found to a working AI stack

On March 16, 2026, at 10 PM in Kuala Lumpur, I typed a command into my MacBook's terminal for the first time in my life.

brew install ollama

The terminal returned: command not found.

I didn't even have Homebrew installed. I didn't know what a package manager was. I had never opened Terminal before that night. For 17 years, my tools were employment contracts, competency frameworks, and performance management systems — not command lines.

48 hours later

I had a full local AI stack running on my MacBook:

• Six AI models: qwen2.5:14b, qwen2.5-coder:14b, deepseek-r1:14b, gemma3:12b, llava:13b, nomic-embed-text
• ChatGPT-style interface: Open WebUI running on localhost:3000
• Coding assistant: VS Code + Continue with qwen2.5-coder
• RAG system: AnythingLLM with LanceDB, chunking at 1000/200 overlap
• Cross-device VPN: Tailscale connecting MacBook and Mac Mini
• Vision model: Gemma 3:12B for image understanding

Two days. From command not found to a working AI system.

The RAG test that changed everything

I uploaded a company HR handbook — 4.3MB, dense with policy language — into the RAG system. Then I asked it a question about something that wasn't in the document.

The AI said: "The information is not available in the provided context."

No hallucination. No fabrication. No confident nonsense.

That sentence — that refusal to guess — was the most important sentence an AI has ever said to me. Because in HR, wrong answers destroy trust. If I tell an employee their medical leave entitlement is 60 days when the handbook says 14, I have created a liability, not a solution.

Key technical findings

1. RAG hallucination test PASSED — HR Handbook query returned "not in context" for missing info
2. Gemma 3:12B confirmed as preferred vision model over LLaVA:13B (LLaVA hallucinated a car dashboard from a macOS screenshot)
3. Chunking: 1000 chars / 200 overlap works for HR handbooks
4. AnythingLLM "Embed File" is the correct RAG path — "Continue Anyway" stuffs context and fails for large docs

What this means

That is not a statement about talent. That is a statement about timing. The tools have matured to the point where a non-engineer can build what would have required a team of specialists five years ago.

And that changes everything — not just for me, but for every domain expert sitting on years of irreplaceable knowledge.

The gap is not technical. The gap is not domain knowledge. The gap is the bridge between the two.