Study plan — AI-103
A suggested two-week ramp. Domains 1–2 carry the most weight (55–65% combined), so spend the most time on Foundry models/agents, RAG, and orchestration.
Week 1 — Build the foundation
| Day | Focus | Practice |
|---|---|---|
| 1 | Read Domain 1: choosing Foundry models & services, deployments | — |
| 2 | Quotas/scaling/cost, monitoring, security, responsible AI | Domain 1 quiz |
| 3 | Read Domain 2: generative apps, RAG, SDKs | — |
| 4 | Agents: tools, function-calling, memory, multi-agent orchestration, observability | Domain 2 quiz |
| 5 | Read Domain 3: image/video generation, multimodal understanding | Domain 3 quiz |
| 6 | Read Domain 4: text analysis, speech | Domain 4 quiz |
| 7 | Read Domain 5: retrieval/grounding, document extraction · drill flashcards | Flashcards |
Week 2 — Reinforce and simulate
| Day | Focus | Practice |
|---|---|---|
| 8 | Re-take any quiz scored below 80% | Quizzes |
| 9 | Hands-on: build a Foundry RAG app over Azure AI Search | Labs |
| 10 | Hands-on: build a Foundry agent with tools + function-calling | Labs |
| 11 | Multiple-response set and case studies | Practice |
| 12 | Full mock exam, review every miss | Mock exam |
| 13 | Targeted re-study of the lowest-scoring domain | Flashcards |
| 14 | Second mock exam attempt · light review | Mock exam |
Tips for exam day
- Over-invest in Domains 1–2 — they're the majority of the exam.
- For "which model/service" questions, match the task: LLM vs small vs multimodal, and the right Foundry Tool for grounding, vision, speech, or extraction.
- For agents, know tools/function-calling, memory, orchestration, and approval/oversight patterns cold.
- For "most secure" questions, favor managed identity + RBAC + keyless credentials and private networking.
- For responsible AI, know content filters, evaluators/safety evaluations, tracing, and provenance/audit.