Azure AI-901 · Module 7 · Session 11
Session 11: Mock Exam 1 & Gap Review
Sit the first timed, full-length mock exam, review every question with its reasoning, analyse your score by domain and skill, and revise the weak areas: both the concepts and the Foundry hands-on steps behind them.
▶ Start here: open the AI-901 Exam Practice page and choose Mock Exam 1 (40 questions, 45 minutes). Use timed mode with no notes, exactly like the real exam.
Before you start
- Use a quiet 45-minute slot. Close other tabs and put your phone away.
- Do not pause the timer. On the real exam you cannot.
- Answer every question; there is no negative marking. Flag doubtful questions and return to them if time allows.
How to review a mock (the most important hour of the course)
For every question, right or wrong, ask three things:
- Why is the right answer right? Say it in one sentence using the exam’s vocabulary (“sentiment analysis returns confidence scores…”).
- Why is each wrong option wrong? Most AI-901 distractors are real Azure features used in the wrong place (for example OCR offered for a sentiment scenario).
- Was it a knowledge gap or a reading mistake? Reading mistakes (missing “NOT”, “most cost-effective”, “minimise development effort”) are fixed with technique; knowledge gaps need revision.
Domain-wise score analysis
The practice page shows your score per domain and per skill. Fill this table after the mock:
| Skill area | Questions | Correct | % | Action |
|---|---|---|---|---|
| Responsible AI principles | ||||
| Model concepts, choice and parameters | ||||
| AI workloads: generative, agentic, text, speech, vision, extraction | ||||
| Foundry apps: prompts, deployments, chat SDK | ||||
| Agents | ||||
| Text and speech solutions | ||||
| Vision and image generation | ||||
| Content Understanding |
Below 70% in an area is a priority. Domain 2 counts for 55–60% of the real exam, so weakness there costs the most.
Revision map: weak area → what to redo
| If you missed… | Revise | Redo |
|---|---|---|
| Responsible AI scenarios | Session 2 trigger-word table | Lab 1, Step 6 worksheet |
| Temperature, top-p, max tokens, deployment types | Session 2 and Session 5 | Lab 1, Step 4 |
| Choosing a workload | Sessions 3 and 4 scenario drill | Lab 2, Step 5 |
SDK code (AIProjectClient, get_openai_client, responses.create) |
Session 6 | Lab 3, Steps 2–3 |
| Agents and tools | Session 7 | Lab 3, Steps 4–5 |
| Language and Speech SDK calls | Session 8 | Lab 4 |
| Multimodal image input and image generation | Session 9 | Lab 5 |
| Analyzers, field schemas, async results | Session 10 | Lab 6 |
Most common gaps after Mock 1
- Confusing transparency with accountability.
- Picking object detection when the question only needs classification (or the reverse).
- Forgetting that code calls the deployment name.
- Mixing up the project endpoint (
/api/projects/<name>, for the Foundry SDK and agents) with the resource endpoint (for Language, Speech and Content Understanding). - Not knowing which tool an agent needs: file search (documents) vs code interpreter (calculations and data files) vs function calling (your own actions).
- Thinking a chat model remembers previous calls without history.
Check yourself
Your Mock 1 score is 64% overall, but 45% in Content Understanding. What do you do first?
Re-read Session 10, redo Lab 6 (analyzers, schema, polling the result) and then retake the Content Understanding topic drill on the practice page.Why review the questions you answered correctly?
Some were lucky guesses. Explaining why each option is right or wrong turns a guess into reliable knowledge.Created by Ajeetkumar