A certification proves knowledge. Labs prove ability.
Six labs, each one a working thing you build — mapped to what the Claude Developer exam tests and to what production AI work looks like. Included in Guided + Labs ($199) and the Production AI Engineer Programme ($999).
Build with the Claude API
You build: A small application using messages, structured output and tool use, with sensible error handling.
- API integration
- Tool use
- Structured output
- Model selection
Connect an agent to real systems with MCP
You build: An MCP server that exposes a real data source and actions to an agent, with permissions you'd accept in production.
- Model Context Protocol
- Tool design
- Least-privilege access
Agentic development with Claude Code
You build: A real feature shipped in an existing repository using Claude Code — planning, subagents, review and tests.
- Agentic SDLC
- Context management
- Review discipline
RAG that you can measure
You build: A retrieval pipeline over a real document set, with retrieval quality measured rather than eyeballed.
- Chunking and retrieval
- Context engineering
- Grounding
Evals and debugging
You build: An evaluation harness for an LLM feature, used to find and fix a real regression.
- Test sets
- Graders
- Regression tracking
- Debugging model behaviour
LLMOps in production
You build: Observability, cost tracking and prompt caching added to a running service, with a before-and-after cost report.
- Tracing
- Token and cost control
- Caching
- Rollback
With a trainer, not just a worksheet.
Each lab has a brief, a starter repository and a review. Live sessions run over weekends, so you can do this alongside a full-time job. OpenAI and cloud deployment appear alongside Claude where it matters for the work.
One message when the first lab cohort has dates. Nothing else.
Take the check first. It points you at a track.
Free, no card, and scored by exam area.