Curriculum
Twelve weeks. Three phases.Free.
Vibe coding → engineering depth → track specialization. Each module pairs curated external YouTube videos for the fundamentals with a Pandai SHIP project for the practice. Same spine in Pandai Open (free) and the in-person cohort.
How the curriculum is structured
Curated externals. Pandai-built layer on top.
We don’t reinvent fundamentals — we curate the best external explanations and stack the Pandai-specific work on top: SHIP projects, Indonesian context, AI-Native code review, and the operator toolkit you actually use day-to-day.
External fundamentals
Each module starts with curated YouTube videos for the underlying concepts — Karpathy, the Anthropic team, others who already explained this stuff well. We don’t reinvent what the field already taught.
Pandai layer
SHIP project. Indonesian context. Code review against AI-Native discipline. This is where the Pandai work happens.
Operator toolkit
Slides. Timelines. Spreadsheets. Images. Admin docs. All built with AI as you go — because engineers ship more than code.
Pre-week (W0)
M0 — Set up your engineer brain
Async qualifier before Week 1. Python or TypeScript fluency. Git basics. Docker basics. API basics. Cursor + Claude. If you can’t clear M0, the rest of the curriculum will be painful; we’d rather you spend a few extra weeks here than skip it.
Three phases
AI Vibe Coding → Engineering Depth → Track Specialization.
The 12 weeks split into three phases. Each phase is a coherent block; you can dip in mid-program if you have prior context, but the intended path is straight through.
AI Vibe Coding
Ship something useful with Claude. By Friday of Week 4 you’ve built an agent that does real work — not a toy. The on-ramp for every track.
Week 1 · M1
Claude Code as your dev environment
SHIP: A real-annoyance CLI built and shipped from a fresh terminal. Roommate can install it.
Week 2 · M2
LLM fundamentals — RLM first
SHIP: Token + cost calculator across Claude / Gemini / OpenAI for an Indonesian-language workload. Long-context vs chunked variants compared.
Week 3 · M3
Agents and MCP
SHIP: A 3-tool agent built end-to-end from off-the-shelf MCPs. (Writing your own MCP comes in W4.)
Week 4 · M4
Practical agent shipping
SHIP: Pick one and ship it: video (Remotion + Claude), slides (Sheets/Slides MCP), Excel reconciliation, Indonesian forms (KTP / NPWP / tax), or a website-registration agent.
Engineering Depth
From toy to product. From solo coder to AI-orchestrator. Auth, schema, eval harness, observability, deploy — with the Bella architecture as the through-line. Then five days of track exposure before you pick.
Week 5 · M5
Bella case study
SHIP: A re-implemented Bella-style auth + scheduler slice with your own eval harness and one observability dashboard.
Week 6 · M6
Become the CEO of your codebase
SHIP: A feature shipped end-to-end using the Think → Plan → Build → Review → Ship → Reflect loop, with Claude Code subagents playing roles you used to do alone. Observability dashboard included.
Week 7 · M7
Pick your track
SHIP: A 1-page track-rank memo. Pick your track Friday based on five 1-day deep dives.
Track Specialization
Pick your specialty. Five tracks — AI Engineer, Software, Data, Quality, Tech Ops. The AI Engineer track ends in verifiable closed-loop systems (autoresearch), not fine-tuning. Every track ships a capstone.
Week 8 · M8
Track depth begins — evals before everything
SHIP: AI Engineer: an eval harness over your W4 agent, with metrics that matter for your use case. Other tracks: first deep-dive project in your specialty.
Week 9 · M9
Build your first closed-loop system
SHIP: AI Engineer: an LLM-wiki — a system that researches questions, grades its own answers via your ontology, and updates what it knows. Cohort: Gate 2 placements begin for the next 40%.
Week 10 · M10
RAG, finally — for what RLM can’t carry
SHIP: AI Engineer: a RAG slice (chunking, pruning, GraphRAG over your ontology) motivated by a cost or freshness ceiling you measured. Others: integration + observability project on your prior SHIP.
Week 11 · M11
Self-improving agents in production
SHIP: Real-client capstone build. AI Engineer: a verifiable closed-loop — the agent runs, grades its own output, and is measurably better Friday than Monday. MLOps essentials wired in: serving, prompt versioning, token analytics, human-in-loop.
Week 12 · M12
Demo Day · Gate 3
SHIP: Capstone demoed live to the cohort. AI Engineer: self-improving autoresearch — the system gets better while the audience watches. Plus a final pass on go-to-market thinking. Top performers join Metatech as FDEs.
Week by week
The full schedule.
One module per week. SHIP project at the end of each. Public GitHub repo for everything you build.
- W0 (pre-week)M0
Set up your engineer brain
SHIP:Python/TS, Git, Docker, API basics, Claude Code installed. AI woven into daily life.
- W1M1
Claude Code as your dev environment
SHIP:Real-annoyance CLI (using AI → bending AI)
- W2M2
LLM fundamentals — RLM first
SHIP:Token + cost calculator (RLM vs chunked)
- W3M3
Agents and MCP
SHIP:3-tool agent with your own MCP (automation → agentic-AI arc)
- W4M4
Practical agent shipping
SHIP:Video / slides / Excel / Indonesian forms / website registration
- W5M5
Bella case study
SHIP:Auth + scheduler slice + eval harness + 100/40 memory rule applied
- W6M6
Become the CEO of your codebase
SHIP:Feature shipped via Think→Plan→Build→Review→Ship→Reflect loop
- W7M7
Pick your track
SHIP:Track-rank memo after five deep dives
- W8M8
Track depth — evals before everything
SHIP:AI Eng: eval harness over W4 agent. Others: first deep-dive project
- W9M9
Build your first closed-loop system
SHIP:AI Eng: LLM-wiki (self-grading researcher). Cohort: Gate 2 placements
- W10M10
RAG, finally — for what RLM can’t carry
SHIP:AI Eng: RAG slice + GraphRAG over your ontology. Others: integration project
- W11M11
Self-improving agents in production
SHIP:Real-client build. AI Eng: closed-loop, measurably better by Friday
- W12M12
Demo Day + Gate 3
SHIP:Capstone demoed live. AI Eng: self-improving autoresearch + GTM thinking
Two ways through
Take it free, async. Or take it in-person, with placement.
Same modules either way. The free version is yours to take at your pace. The cohort version adds the operational simulator (The Yard), the real-client capstone, and the placement pipeline.