Building Private Chat w/ Docs Using Claude Code (Start to Finish) — Lesson 4

The Personal AI Architecture
1

The full process live — interview, spec, build plan, execute, test — to a working personal AI system on my own machine. No internet connection needed.

In Lesson 4 I walk through the 5-step process end to end and end up with a working chat-with-documents app running fully locally.

What you’ll see

  • The interview — the AI coding agent kicks off, asks me what I want to build, who it’s for, document types, conversation history, what happens when files change. I tell it to keep things as simple as possible for V1 — one brick at a time
  • The spec — written with proper user stories + acceptance criteria so the AI can build against them and test against them. Time spent on the spec is time saved later
  • The build plan — phases with verification at the end of each, written so a non-coder can keep going
  • Execute — the AI runs the build plan ~20 minutes, phase by phase. At one phase, 40 of 42 tests pass — it finds the failures and resolves them itself before moving on
  • Test — open the local app in the browser, ask “What is BrainDrive?”, get a real answer from the indexed documents

A few opinions baked into this lesson

  • One brick at a time. AI coding agents are eager to add features; the simple-first ethos is a feature, not a constraint
  • Skip the vector DB. A good table of contents the AI builds itself + agentic crawling is simpler and (in my experience) more powerful for personal-scale doc corpora
  • Local-only. Ollama running qwen2.5 (good tool-calling support); local web app in the browser; no internet connection needed

At the end you have a working personal AI system — 4 components + 2 APIs running on your machine, with a model, interface, and tools wired up. A foundation to build whatever you want on top of, and the process to do it.

Resources

Discussion

Build the chat-with-docs app from the template? Share what you built — or which step was the trickiest. Where the loop fired for you is the most interesting part.

Previous: Lesson 3 — 5 Step Workflow for Building Your AI System. Next: Lesson 5 — Using BrainDrive as Your Personal AI System (MIT Licensed).

5 Step Workflow for Building Your AI System — Lesson 3