â›Šī¸ Pedagogical Engineering • Moving Beyond the Chatbox

From Passive Chatting to
Disciplined AI Orchestration

Master real-world AI agent architecture through engineering rigor. Learn to design, deploy, and debug autonomous workflows using the battle-tested 5D Method, progressive Belt curriculum, and immutable source verification.

5D
Engineering Framework
5 Belts
Step-by-Step Progression
Sealed
Sources You Can Check
5+ Agents
Tools You Set Up Yourself
Live Orchestration Architecture

The Anatomy of an Autonomous Agent

Watch a request travel the pipeline: planned, executed with tools, and checked by a verifier before anything is delivered. Nothing reaches the output without passing the check.

⚡ Dojo Multi-Agent Execution Visualizer
[10:00:00.000] [SYSTEM] Agentic runtime initialized. Awaiting pipeline execution trigger...
The Core Framework

The 5D Pedagogical Method

Every project in the Dojo moves through five structured phases to instill critical judgment over what AI models generate.

Phase 01
D1: Discussion
Phase 02
D2: Design
Phase 03
D3: Develop
Phase 04
D4: Deploy
Phase 05
D5: Debug

đŸ—Ŗī¸ D1: Discussion — Scoping & Constraints

Before touching any code or prompts, define the exact problem boundary. Let the AI interview you to surface hidden assumptions, spot the information you have not supplied yet, and head off made-up answers before they start.

Pedagogical Rule: "Prompting without scoping produces brittle illusions. Scoping turns AI into a predictable engineering multiplier."
PHASE_01_SPEC.md Verified Spec
# Architectural Scope: D1 Discussion
- Objective: US Retirement Penalty Validator
- Corpus Boundary: IRS Pub 590-B + CA FTB Form 3805P
- Verification Gate: SHA-256 hash match required
- Strict Rule: Zero ungrounded tax advice generation
Step-by-Step Progression

The Progressive Belt Curriculum

A systematic 5-tier mastery ladder taking you from privacy-conscious prompt design to production multi-agent cloud infrastructure.

Phase 1: Foundational Literacy

White Belt — AI Landscape & Cost Literacy

Privacy Settings, Comparing the Big AI Models & What They Really Cost

Build your foundational hygiene. Configure data privacy opt-outs across major frontier LLMs, understand the trade-offs between Claude, GPT-5, and Gemini, and master what each really costs to run.

  • ✓ Enterprise privacy controls & zero-training toggles
  • ✓ What you actually pay: sending, receiving, and reusing text
  • ✓ Letting the AI interview you to pin down what you need
Milestone Capstone Deliverable
Frontier Model ROI Calculator
A personalized configuration sheet and cost model comparing token burn rates across 5 production workflows.
Tooling Ecosystem
ChatGPT Plus Claude Pro Google Gemini Tokenizer CLI
Evidence-Based Learning

Working Capstone Reference Applications

Real-world, production-deployed web tools built using the Dojo's 5D Method, on sealed source documents that are checked before every answer.

LIVE REFERENCE • v2 & v3

Retirement Decumulation Reference

US retirement penalty and early-distribution rules with a California overlay. The same question always gets the same answer, drawn only from a fixed set of IRS and FTB tax publications that are checked for changes before every run.

DUE DILIGENCE

Bay Area City Insight

Home-buying due-diligence engine covering municipal demographics, housing valuations, school quality scores, and public safety data for San Francisco Bay Area cities.

OPTIMIZATION ENGINE

Award Transfer Optimizer

Airline and hotel points transfer partner matrix. Evaluates transfer ratios, sweet spots, and booking alliances across major credit card loyalty programs.

View All Capstone Architecture Specs →
Pedagogical Rigor

Why Pedagogy Beats Prompting

Copy-pasting prompts from the internet gives a false sense of security. The Dojo teaches fundamental mental models so you can architect resilient systems.

🧠

Let the AI Interview You

Instead of demanding quick answers, teach AI to interview you. Force the model to elicit edge cases, operational boundaries, and formatting constraints before generation.

🔍

"Verify, Don't Trust"

Learn systematic debugging: log parsing, automated assertions, and source-document fingerprinting — so you can prove the evidence an agent quoted is the evidence you actually gave it. You are trained to spot subtle LLM hallucinations immediately.

⚡

Two Tracks, Your Choice

Choose your pathway: Track A (Browser-First) using Claude Cowork and GUI tools, or Track B (Terminal-First) using Claude Code, bash automation, and background jobs on your own Linux server.

đŸĨ‹
Mike Cho
Senior Project Engineer & AI Coach
Silicon Valley Engineering Heritage

Engineering Discipline Meets Modern AI

With over 20 years of experience in hardware validation, optical test engineering, and network protocol validation in Silicon Valley, Mike brings real industrial rigor to AI development. The curriculum is designed to transform enthusiastic learners into people who confidently direct AI to build software they can actually check.

đŸ‡ē🇸 English 🇹đŸ‡ŧ 國čĒž / æ™Žé€ščŠą 🇭🇰 į˛ĩčĒž (Cantonese)