Build a production LangGraph agent from scratch in 6 weeks. Study a real architecture, replicate it for your own domain, ship it live. AI reviews your code. Mentors step in when it matters.
In partnership with Moonstone Labs — top performers get invited to work on real client projects
You'll replicate the architecture of a production agent service, then rebuild it for a domain you choose: StudentBuddy, JobBuddy, FitnessBuddy, or your own idea.
Agent graph architecture
START → initialize → router → generate_reply → generate_decision
↑ ↓
└──────── memory_updater ───────┘
↓
implementation → ENDDefine your agent's domain, sections, and acceptance criteria. Write a product spec with user stories and edge cases. MOSS reviews your doc, then conducts a live video interview to pressure-test your thinking before you write a line of code.
Produce an architecture diagram, LangGraph state schema, and Supabase table definitions. MOSS challenges you on one key design decision — caching strategy, memory model, or routing logic — in a second video review before sign-off.
6 pull requests, one node at a time: initialize → router → generate_reply → generate_decision → memory_updater → implementation. Every PR must include a LangSmith trace and a written tradeoff note. MOSS reviews each PR in Slack within minutes.
Build a golden dataset of 20+ test cases covering happy path, edge cases, and adversarial inputs in LangSmith. Run evals until your score hits ≥ 0.8 across all sections. MOSS runs a final adversarial sweep before sign-off.
Ship backend to Railway, frontend to Vercel. Stand up a LangSmith monitoring dashboard with live traces, latency metrics, and error rates. Once metrics are stable for 48 hours, a mentor joins a final video call to sign off. You graduated.
Designed with Moonstone Labs, an AI studio that ships production agent systems for growth-stage companies. The curriculum mirrors real client work — and strong graduates get invited to interview for paid projects.
Talk through your architecture live with an AI interviewer. No slides, no docs — just a conversation, like a real engineering team.
Every PR gets structured AI feedback: what's good, what needs work, and the tradeoff questions a senior engineer would ask.
PR discipline, LangSmith traces, tradeoff reasoning, production monitoring. The habits that get you hired.
You don't get a PDF. You get a live credential — a URL employers can open and interrogate.
Every graduate gets a unique URL tied to their actual work — PRs, eval scores, LangSmith traces, and design reviews. Share it in a job application or on a resume. Employers get the real picture, not a summary.
Move your cursor over the cardMOSS connects to ChatGPT and Claude via MCP. Scan the QR code and the candidate's full work record loads directly into the employer's AI tool — ready to answer any question about skills, decisions, or position fit.
“We designed this curriculum the same way we scope client work — start from a real system, understand why it's built that way, then rebuild it. That process produces engineers who think, not just engineers who prompt. Those are the people we want on hard problems.”
Top performers are invited to interview for paid client projects at Moonstone Labs — working on real production systems alongside the team that designed this curriculum.
Applications reviewed within 2 business days.
