The world's first coding-agent-native learning platform.
Learn AI architecture by building with an AI architect — inside your terminal.
Traditional course: Read docs → Watch videos → Maybe build something
AI Architect Academy: Clone repo → Open Claude Code → Build with an instructor
When you open this repo in Claude Code, the agent becomes your personal AI architecture instructor. It knows your progress, guides you with questions (not answers), and reviews your work like a senior engineer.
The medium is the message. You learn to build AI systems by building AI systems — with an AI.
# 1. Clone the academy
git clone https://github.com/frankxai/ai-architect-academy.git
cd ai-architect-academy
# 2. Open Claude Code
claude
# 3. Start learning
/academyThat's it. The instructor takes over from here.
When you launch Claude Code in this directory, you'll see:
Academy | 0/3 labs completed | Current: none | Type /academy for menu
Type /start-lab 01 and the instructor begins:
╔══════════════════════════════════════════════════════════════╗
║ AI ARCHITECT ACADEMY — Lab Mode ║
║ Lab: Fix the Broken RAG Pipeline ║
║ Difficulty: Intermediate | Est. Time: 45m ║
╚══════════════════════════════════════════════════════════════╝
Welcome, Architect.
Your company's customer support AI is returning irrelevant answers.
The RAG pipeline is live but broken. There are 3 bugs in rag_pipeline.py.
Before I guide you — take a look at the code. What do you notice about
the chunk_documents() method?
The instructor uses Socratic method — it asks questions, not gives answers. When you're stuck, type /hint for incremental clues. When ready, type /review for an architect-grade code review.
| Lab | What You Build | Difficulty | Time |
|---|---|---|---|
| 01: Fix the Broken RAG Pipeline | Debug chunking, search, and context assembly in a production RAG system | Intermediate | 45m |
| 02: Build a Multi-Agent System | Implement a coordinator that orchestrates Researcher, Analyst, and Writer agents | Advanced | 60m |
| 03: Build Your Own MCP Server | Create a TypeScript MCP server with 3 tools for Claude Code | Advanced | 60m |
Each lab includes:
- Real source code (broken or scaffolded)
- Test suites (your objective pass/fail gate)
- Sample data
- Checkpoints tracked in your progress file
| Command | What It Does |
|---|---|
/academy |
Main menu — see everything available |
/start-lab 01 |
Begin (or resume) an interactive lab |
/hint |
Get a Socratic hint — direction, not answers |
/review |
Architect review: tests + score + feedback |
/next |
Context-aware "what should I do next?" |
/progress |
Your full progress dashboard |
/solution |
Reveal the solution (requires prior attempt) |
| Command | What It Does |
|---|---|
/design-solution |
End-to-end AI solution architecture |
/build-rag |
Build a RAG system from scratch |
/mcp-server |
Build a custom MCP server |
/security-review |
Security assessment |
/optimize-costs |
Cost optimization analysis |
ai-architect-academy/
├── CLAUDE.md # Instructor Engine — persona, rules, curriculum state
├── .academy/
│ └── progress.json # Your progress (gitignored, personal to you)
├── .claude/
│ ├── commands/ # /hint, /review, /next, /academy, etc.
│ ├── hooks/ # Session-start welcome, progress tracking
│ └── skill-rules.json # Auto-activation rules for 80+ skills
├── labs/ # Interactive coding labs
│ ├── 01-rag-pipeline/ # Python — fix broken RAG (3 bugs)
│ ├── 02-multi-agent-system/ # Python — build coordinator pattern
│ └── 03-mcp-server/ # TypeScript — build MCP server
├── claude-ai-architect/ # Knowledge base + 22 expert skills
│ ├── skills/ # RAG, MCP, multi-cloud, security, etc.
│ ├── knowledge-base/ # OCI GenAI, infrastructure docs
│ ├── templates/ # D2 diagrams, Terraform modules
│ └── saas-curriculum/ # 12-week structured curriculum
├── 01-design-patterns/ # 20+ architecture patterns
├── 02-learning-paths/ # Structured learning tracks
└── 05-projects/ # 100+ project ideas
-
Instructor Engine (
CLAUDE.md) — Defines the teaching persona, Socratic rules, skill activation, and progress tracking. This is what turns Claude from a generic assistant into a domain-expert instructor. -
Interactive Labs (
labs/) — Real codebases with real bugs and real test suites. The student writes code. The instructor guides. Tests are the judge. -
Knowledge Base (
claude-ai-architect/skills/) — 22+ deep-dive skills that auto-activate based on context. Working on RAG? The RAG skill loads. Building an MCP server? The MCP patterns load.
| Path | Hours | Focus |
|---|---|---|
| Foundation | 20h | Claude SDK, MCP basics, first agent |
| Agent Developer | 40h | Multi-agent, RAG, orchestration |
| Multi-Cloud | 40h | OCI, AWS, Azure, GCP patterns |
| Enterprise Lead | 30h | Security, governance, compliance |
| Bootcamp | 40h | Intensive: all labs + capstone |
| Level | Requirements |
|---|---|
| Associate | Complete 2 labs with avg score >= 70 |
| Professional | Complete all 3 labs + 2 learning path modules |
| Expert | All labs (avg >= 85) + capstone project + security review |
Auto-activated by context. You never need to load them manually.
| Category | Count | Examples |
|---|---|---|
| Agent Frameworks | 12 | Claude SDK, Oracle ADK, LangGraph, OpenAI Agents, CrewAI |
| Multi-Cloud AI | 16 | OCI, AWS Bedrock, Azure OpenAI, Vertex AI, Terraform |
| MCP & Integration | 10 | MCP architecture, server builder, tool use patterns |
| RAG & Knowledge | 8 | Vector DBs, chunking, retrieval strategies, knowledge graphs |
| Enterprise & Security | 12 | Guardrails, compliance, audit logging, red teaming |
| Evaluation & Production | 18 | LangFuse, LangSmith, CI/CD, monitoring, scaling |
- Claude Code — The agent runtime
- Model Context Protocol — Tool integration standard
- ACOS patterns — Skill activation, hooks, and progress tracking adapted from the Agentic Creator OS
PRs welcome. If you want to add a lab, create a labs/[id]-[name]/ directory with:
.lab/config.json— Lab metadata- Source code (broken or scaffolded)
tests/— Test suiteREADME.md— Mission briefing
Built by FrankX
The era of watching videos to learn code is over. Build with agents.