AI-901 FDP · Day 6 · All decks
DAY 6 · 10-Jul

Rapid Revision, Exam Traps
& Mock Certification Test

Decision tree — burn this in

flowchart TD
  S([Scenario]) --> Q1{Create content?}
  Q1 -->|text| T[Generative model]
  Q1 -->|image| IG[Image-gen]
  Q1 -->|video| VG[Video-gen]
  Q1 -->|no| Q2{Act w/ tools?}
  Q2 -->|yes| AG[Agent]
  Q2 -->|no| Q3{Modality}
  Q3 -->|image/audio in| MM[Multimodal]
  Q3 -->|audio to text| STT[Speech recognition]
  Q3 -->|text to audio| TTS[Speech synthesis]
  Q3 -->|analyze text| TA[Text analysis]
  Q3 -->|fields from files| CU[Content Understanding]

Top exam traps

  1. temperature/top_p = randomness · max_tokens = length
  2. Reliability&Safety (system) ≠ Accountability (people)
  3. Code references the deployment name, not base model
  4. Rules/grounding → system message
  5. Multimodal understands · image/video-gen create
  6. Agent uses tools & acts; chat = text-in/text-out
  7. Structured extraction from files → Content Understanding + field schema
  8. Grounding/RAG fixes hallucination (not higher temperature)

Memory sheet — concepts

F.R.I.T.A.P.

Fairness · Reliability&Safety · Inclusiveness · Transparency · Accountability · Privacy&Security

GenAI

token · embedding · transformer · predict-next-token · grounding vs hallucination · multimodal

Config

temp/top_p=random · max_tokens=length · penalties=repetition · seed=repeatable

Deploy

Standard · Global Standard · PTU(reserved) · Batch · Serverless

Memory sheet — Foundry

Structure

Hub → Project → Deployment (name+endpoint+filter)

Chat

messages: system/user/assistant → choices[0].message.content

Agent

model+instructions+tools; agent→thread→message→run→read

Content Understanding

analyzer + field schema; docs/images/audio/video

Lightning round

"People are answerable + a governance board oversees it" → principle?
Accountability = people/orgs answerable + governance.

Lightning round

Reduce hallucinations in answers about your internal policies?
Grounding/RAG supplies retrieved facts so answers are based on your data.

Lightning round

Minimum to create & test a single agent in the portal?
An agent needs a model deployment + instructions; tools/knowledge are optional add-ons.

Mock Certification Test

50 questions · 60 minutes · closed book

Weighted 40/60 across the two domains. Pass ≈ 35/50.

  • Read "choose TWO/THREE" carefully
  • Name the capability first, then the Foundry way to build it
  • Flag & return to hard items — don't stall
  • Sandbox the exam UI first: aka.ms/examdemo

Then review every answer together.

You're ready. 🎯

Register with a personal Microsoft account · pass = 700/1000 · cert never expires.