Challenges¶
Challenges are less guided than labs. Use them after completing the related labs to prove you can apply the concepts without a step-by-step recipe.
| Challenge | Title | Level | Focus | Time |
|---|---|---|---|---|
| Challenge 001 | Build an Agent Loop from Scratch | L100 | Agent loop, tools, state | ~60 min |
| Challenge 002 | Fix a Broken RAG Pipeline | L200 | RAG, retrieval, evaluation | ~75 min |
| Challenge 003 | Defend an Agent from Prompt Injection | L200 | Security, context filtering | ~60 min |
| Challenge 004 | Optimize a Token Budget | L200 | Cost, context, prompt budget | ~60 min |
| Challenge 005 | Build a Safe MCP-Style Tool | L200 | MCP, schema, privacy | ~60 min |
| Challenge 006 | Investigate an Agent Observability Incident | L300 | Tracing, metrics, root cause | ~60 min |
| Challenge 007 | Red-Team Agent Guardrails | L300 | Guardrails, PII, safety | ~60 min |
| Challenge 008 | Build a Local-Only Agent | L200 | Local tools, routing, no cloud | ~75 min |
How Challenges Work¶
- You get a scenario, objective, constraints, starter files, and rubric.
- You do not get a step-by-step walkthrough.
- When possible, a challenge includes a local validator that generates a completion code.
- Enter the code on the challenge page to confirm the final result.