AI Learning Hub
medium

Medium-track milestone quiz

Twelve questions covering embeddings, vector DBs, RAG, prompt engineering, tools, agents, evals, and caching. Pass with ≥9/12.

The whole stack in one picture

A production RAG agent — every piece you've just learned, in order:

Every concept from the medium track, wired together as one production request path.

Check your understanding

  1. 1. In RAG, what does the 'retrieve' step actually do?
  2. 2. Why is metadata filtering essential in multi-tenant vector search?
  3. 3. Most reliable way to get strict JSON from an LLM:
  4. 4. Few-shot prompting is most useful when:
  5. 5. Tool calling: who actually runs the tool?
  6. 6. Anthropic's 'Building Effective Agents' argues that:
  7. 7. What does `cache_control: ephemeral` on a static prefix do?
  8. 8. LLM-as-judge evals are best for:
  9. 9. Why is hybrid search usually better than pure vector?
  10. 10. Building a flight-booking assistant — safest way to expose the booking action:
  11. 11. Right chunks retrieved, but the model answers from training data anyway. Fix:
  12. 12. Cost is dominated by re-sending a 5k-token system prompt every turn. Best fix:

What's next

You're ready for the Advanced track: how transformers actually work, the training stages that produce a chat model, when to fine-tune, multi-agent systems, MCP, evals at scale, multimodal models, safety, and the open-source ecosystem.

Related lessons in this track