Lab 7: Mock Exam Analysis

Turn two mock exams into a personal revision plan: attempt Mock Exams 1 and 2 under timed conditions, log every wrong answer with the skill it tests, redo the related lab step for each weak skill, and build your revision checklist.

Time: 2–3 hours · You need: the AI-901 Exam Practice page, and a spreadsheet or notebook.

Outcome. A clear, personal list of what to revise before the exam.

Step 1: Attempt both mocks under timed conditions

1. Open Mock Exam 1 in exam mode (45 minutes, no feedback until the end).
2. Take a 10-minute break, then Mock Exam 2.
3. Note both scores. The practice page saves your best score per mock on this device.

Step 2: Log every wrong answer

Use the review screen after each mock. Copy each wrong (or guessed) answer into a log:

Pattern check: if more than a third of your errors are misreads, slow down on the question stem; if they cluster in one skill, that skill is your priority.

Step 3: Redo the lab step for each weak skill

Hands-on repetition is the fastest fix for Domain 2 gaps. For each skill in your log, redo the listed step without looking at the solution, then compare:

• Chat SDK → write the AIProjectClient → get_openai_client() → responses.create() script from memory (Lab 3).
• Agents → create a new agent version with different instructions and a file search tool; call it with a conversation (Lab 3).
• Language / Speech → recognise one WAV file and analyse its sentiment (Lab 4).
• Vision → send one image with an input_image part (Lab 5).
• Content Understanding → analyze one invoice and read TotalAmount from the result (Lab 6).

Step 4: Retest the weak topics

On the practice page, open Topic practice, choose each weak skill and answer the drill in study mode (instant explanations) until you reach 80%+ on that topic.

Step 5: Build your revision checklist

Create a one-page checklist for the last week, for example:

• [ ] Six responsible AI principles with one scenario each, from memory
• [ ] Temperature vs top-p vs max tokens; reasoning models use reasoning effort instead
• [ ] Deployment types: Global Standard, Data Zone, Standard, Provisioned, Batch
• [ ] Project endpoint vs resource endpoint, and keys vs Entra ID
• [ ] Agent = model + instructions + tools + knowledge; conversations keep history
• [ ] Language methods: sentiment, key phrases, entities, PII, language detection, summarisation
• [ ] Speech: SpeechConfig, recognizer, synthesizer, voice name, SSML, output format
• [ ] input_image for vision; images.generate (or the image_generation tool) for new images
• [ ] Content Understanding: analyzer, field schema (extract / generate / classify), async polling, confidence

Check yourself