25-Day Programme · 30 Minutes Per Day
WEEK 5 OF 5 · SESSIONS 21–25 · WORK AT YOUR OWN PACE
Domain 2 · Information ExtractionDAY 21
Azure Content Understanding: Extracting From Documents and Forms
Why This Matters
Welcome to your final week. You have come further than most people ever will — four weeks of showing up. This week we finish the syllabus and then we point you straight at the exam. Today’s topic is quietly one of the most employable skills on the whole certificate. Every business you can name — a bank in Lagos, a clinic, a logistics firm — is drowning in paper: invoices, receipts, ID cards, application forms. The person who can teach a computer to read those and pull the numbers out automatically saves a company hundreds of hours. That person gets hired. Today, that starts being you.
30-Minute Module
0:00 – 5:00
What “information extraction” actually means
- Think about a paper invoice. To a human it is obvious: there is a supplier name, a date, a total amount. To a computer it is just a picture — a grid of coloured dots with no idea which part is the “total”.
- Information extraction is the AI workload that turns that messy, human-shaped document into clean, structured data a computer can use — a neat list of fields and values.
- Two words you must separate. Unstructured data is content with no fixed shape: a scanned receipt, a photo, a paragraph of writing. Structured data is organised into named fields, like a row in a spreadsheet:
supplier = "Dangote Cement",total = 45000. - The whole job today is turning the first kind into the second kind. That is the value. A total sitting in a photo helps nobody; a total in a database can be added up, searched and reported.
Whenever you hear “extraction” on the exam, picture an arrow: unstructured document → structured fields. That arrow is the entire workload.
5:00 – 15:00
Azure Content Understanding — the service that does this
- Azure Content Understanding is the service inside Microsoft Foundry built for exactly this: you give it a document, it hands back structured fields. It is the modern, generative-AI successor to the older “Document Intelligence” service — you may see both names, so know they are cousins.
- It works by building an analyzer — a configured “reader” that knows what fields you care about. You point an analyzer at a document and it returns the values.
- There are two ways to get an analyzer. Prebuilt analyzers already know common documents — invoices, receipts, ID documents — so you can use them straight away with zero training. Custom analyzers are ones you teach with a few examples, for a form that is unique to your business.
- Underneath, it still uses OCR — Optical Character Recognition, the technology that reads text out of an image — but Content Understanding goes further: it does not just read the words, it understands which word is the total and which is the date.
- All of this happens inside your own resource group —
rg-yourname-ai901, the one you created on Day 2. Everything this service creates lands there and nowhere else. That is the scope you set up, still doing its job.
Prebuilt = “it already knows this common document”. Custom = “I taught it my own form”. The exam loves that distinction — if the document is a standard invoice or receipt, reach for prebuilt first.
15:00 – 25:00
See it work — the Content Understanding playground
- Go to
ai.azure.com, sign in, and check top-right that you are in the directory holding your subscription — the same check you have done since Day 2. - Open your project, find Content Understanding in the left-hand menu, and open the playground (sometimes labelled “Try it”). A playground is a safe web page for testing a service by hand, before you write any code.
- Choose a prebuilt template — “Invoice” or “Receipt”. Upload a sample document (the portal offers samples, or use a clear photo of a real receipt).
- Watch what comes back: a panel of fields —
MerchantName,TransactionDate,Total— each with the value it found. That panel is structured data. You just turned a photo into a database row without writing a line of code. - Notice each field also has a confidence score — how sure the model is. Low confidence is the model telling you “check this one”. In a real system you would route low-confidence fields to a human. That design pattern is worth an exam mark.
Do not copy-paste anything today — there is no code to type yet. Just click, upload, and read the output carefully. Understanding the shape of what comes back is the whole point.
25:00 – 30:00
Fix it in your own words
- In your AI-901 Notes, write the one-line definition: information extraction turns unstructured documents into structured fields. Then write the difference between a prebuilt and a custom analyzer.
- Note down three fields Content Understanding pulled from your document, and what “confidence” told you about each.
- Then do your assignment. Remember: being stuck is normal, and telling me you are stuck is not failure — it is exactly what a real teammate does. Staying stuck in silence is the only mistake.
Four days of study left after today. You are on the runway now. Keep the pace steady and do not skip.
Key Terms
Information Extraction
The AI workload that turns an unstructured document into structured, named fields a computer can store and search.
Structured vs Unstructured
Unstructured = content with no fixed shape (a scanned receipt). Structured = organised named fields (supplier, date, total).
Content Understanding
The Azure service in Foundry that extracts fields from documents, images, audio and video. The generative successor to Document Intelligence.
Prebuilt vs Custom Analyzer
Prebuilt already knows common documents (invoice, receipt, ID). Custom is one you train on your own unique form.
Today’s Assignment
Run a real document through Content Understanding and report what came back — then explain the workload in your own words.
- Part 1 — do it. In the Content Understanding playground, upload a document (a receipt or invoice — a clear phone photo of a real one is perfect). Use a prebuilt analyzer. Send me a screenshot showing the extracted fields.
- Part 2 — read it. List three fields it extracted, the value for each, and its confidence score. Tell me which field it was least sure about, and why you think that was.
- Part 3 — understand it. In your own words: what is the difference between structured and unstructured data, and when would you choose a custom analyzer over a prebuilt one?
- Part 4 — make it real. Name one Nigerian business or situation where automatic document extraction would save real time or money. One paragraph.
Submit assignment 21 →
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Self-Check — Exam-Style Questions
1. A company scans thousands of supplier invoices and needs the total amount pulled from each one automatically. Which AI workload is this?
- A. Speech recognition
- B. Information extraction
- C. Image generation
- D. Sentiment analysis
B. Information extraction — It is pulling structured fields (the total) out of an unstructured document (the invoice). That arrow — document to fields — is the signature of information extraction.
2. You need to read a standard receipt and get the merchant name, date and total. You have no time to train anything. What should you use?
- A. A custom analyzer you train from scratch
- B. A prebuilt analyzer
- C. Speech synthesis
- D. A large language model with no tools
B. A prebuilt analyzer — Receipts and invoices are common documents Content Understanding already knows, so a prebuilt analyzer works immediately with no training. You only build a custom analyzer when the document is unique to your business.
3. An extracted field comes back with a low confidence score. What is the sensible thing to do in a real application?
- A. Delete the document
- B. Trust it anyway — the model is always right
- C. Route that field to a human to check
- D. Retrain the whole model from zero
C. Route that field to a human to check — A low confidence score is the model honestly telling you it is unsure. A well-designed system uses that signal to send doubtful cases to a person. This “human in the loop” pattern also connects straight back to Responsible AI — reliability and accountability.
Exam Objectives Covered
Information extractionContent UnderstandingStructured vs unstructured dataPrebuilt vs custom analyzersOCRConfidence scores
Domain 2 · Multimodal ExtractionDAY 22
Content Understanding: Images, Audio and Video
Why This Matters
Yesterday you pulled fields out of a document. Today you learn that the exact same service reads far more than paper — it reads photos, listens to audio, and watches video, and pulls structured meaning out of all of them. This is what “multimodal” means, and it is where a lot of the real money is. A company with ten years of recorded customer calls, or a warehouse full of product photos, is sitting on a goldmine it cannot use — until someone extracts it. Learn this and you become the someone. It is also a genuine slice of your exam, so read carefully.
30-Minute Module
0:00 – 6:00
What “multimodal” means — and why it matters
- A modality is just a type of content: text is one modality, images another, audio another, video another. Multimodal means a single service can handle several of them.
- Content Understanding is multimodal. The same service you met yesterday can take an image, an audio file, or a video file and return structured fields — not just documents.
- Why this is powerful: most of the world’s information is not tidy text. It is photos, voice notes, recorded meetings, CCTV. Being able to extract from all of it, with one service and one skill, is a big deal.
- Keep the same mental arrow as yesterday: any input → structured fields. Only the input changes.
If a question mentions photos, audio and documents handled together, the word it is fishing for is “multimodal”.
6:00 – 16:00
The three new modalities, one at a time
- Images. Beyond reading text in a photo, Content Understanding can describe a scene, detect objects, and pull fields from a picture — for example, reading a product label or a whiteboard photographed in a meeting.
- Audio. It first does transcription — turning speech into text (the speech-recognition workload you met in Week 4) — and can then extract structure from that text: who spoke, what was decided, what the customer complained about. It can also do diarisation, which is separating who said what when several people talk.
- Video. Video is really images plus audio over time, so it combines both: transcribing the spoken track, reading text on screen, and describing what is happening in the picture — then returning a structured summary with timestamps.
- The output is always the same shape: named fields and values, ready for a database or a report. That consistency is the whole point of one multimodal service.
New term — diarisation: labelling which speaker said which words. “Speaker 1 / Speaker 2” in a transcript is diarisation at work. Write it in your Notes.
16:00 – 25:00
Try it — extract from an image or a voice note
- Back to
ai.azure.com, your project, Content Understanding, the playground — and check you are in the right directory, as always. - This time pick the image or audio analyzer instead of the document one. For audio, record a short voice note on your phone — say a fake customer message: “Hi, my order 4471 arrived broken, please refund me.”
- Upload it and read what comes back: a transcript of your words, and structured fields the analyzer pulled out — perhaps an order number, an issue type, a requested action.
- For an image, upload a clear photo with text or objects and see how it describes and extracts. Notice again the confidence scores — the same reliability signal as yesterday.
- Everything runs inside your own resource group. Same rule as every day: never share the API key from this service, because it authorises real spending on your subscription.
Do this by hand in the playground today — no code. Tomorrow you will drive this exact idea from Python. Get the concept solid first.
25:00 – 30:00
Lock it in
- Say the four modalities out loud from memory: text, image, audio, video. Then say what “multimodal” means in one sentence.
- Write in your Notes: audio extraction starts with transcription; separating speakers is diarisation; video is images plus audio over time.
- Then do the assignment. Tomorrow is a build day, so make sure today’s ideas are firm — you will lean on them.
Three study days left. You now understand extraction across every kind of content. That is a real, marketable skill — own it.
Key Terms
Modality
A type of content: text, image, audio or video. “Multimodal” means one service handles several of them.
Transcription
Turning spoken audio into written text — the speech-recognition step that audio and video extraction begin with.
Diarisation
Separating who said what in an audio recording, labelling each speaker distinctly (Speaker 1, Speaker 2).
Video Extraction
Combining transcription, on-screen text reading and scene description over time, returning a structured summary with timestamps.
Today’s Assignment
Extract structured information from a non-document — an image or an audio clip — and report the result.
- Part 1 — do it. In the Content Understanding playground, use the image or audio analyzer. Record a short voice note (a pretend customer message) or upload a clear photo. Screenshot the transcript and/or extracted fields.
- Part 2 — explain it. Define, in your own words: modality, multimodal, transcription and diarisation. One clear sentence each.
- Part 3 — connect it. A company has 5,000 recorded customer support calls. Describe, in a short paragraph, how you would use Content Understanding to find out the top three things customers complain about.
- If a modality will not upload or the analyzer errors, tell me exactly what you tried and what the message said. A precise blocker is easy for me to unblock.
Submit assignment 22 →
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Self-Check — Exam-Style Questions
1. A single service can accept documents, images, audio and video and return structured fields from all of them. What word describes this?
- A. Monolithic
- B. Multimodal
- C. Generative
- D. Serverless
B. Multimodal — A modality is a type of content; multimodal means one service handles several types. Content Understanding is multimodal, which is exactly why it can extract from a receipt, a photo and a voice note with the same workflow.
2. When Content Understanding processes an audio recording, what is the first thing it does?
- A. Generates an image of the speaker
- B. Transcribes the speech into text
- C. Translates it into every language
- D. Deletes the audio for privacy
B. Transcribes the speech into text — Audio extraction starts with transcription (speech recognition), turning the spoken words into text that structured fields can then be pulled from. It can also do diarisation — labelling which speaker said what.
3. In a meeting recording, the transcript labels lines as “Speaker 1” and “Speaker 2”. What is this called?
- A. Diarisation
- B. Summarisation
- C. Synthesis
- D. Tokenisation
A. Diarisation — Diarisation separates and labels who said what across an audio recording. Do not confuse it with transcription (speech to text) or summarisation (making text shorter). The exam likes to test these near-neighbours together.
Exam Objectives Covered
Multimodal contentImage extractionAudio transcriptionDiarisationVideo understandingContent Understanding
Domain 2 · Build DayDAY 23
Build a Lightweight Information-Extraction Application
Why This Matters
Today you stop clicking and start building. You are going to write a small Python program that takes a document and extracts its fields — the same thing you did by hand in the playground, but now driven by your own code. This is the difference between someone who has seen a tool and someone who can build with it. When an employer asks “can you actually use this?”, a small working program is the answer that ends the interview in your favour. And you already know the shape of it — it is the four-step pattern you have used since Week 1.
30-Minute Module
0:00 – 6:00
Remember the four-step pattern — it never changes
- Every AI app you have built in this course follows the same four steps. Say them with me: Import → Connect → Send → Use. Today’s app is no different.
- Import — bring in the library that talks to the service. Connect — create a client using your endpoint (the service’s address) and your key (the secret that proves you may use it). Send — hand the service a document. Use — do something with the fields it returns.
- You are not learning a new idea today. You are pointing an idea you already own at a new service. That is exactly how real developers move fast — the pattern carries over.
- Open Google Colab at
colab.research.google.com. Browser only, nothing to install, works fine on a modest laptop.
If you can recite Import → Connect → Send → Use without looking, you are ready. If you cannot, glance at your Week 1 Notes for two minutes first.
6:00 – 16:00
Get your endpoint and key — safely
- In
ai.azure.com, open your Content Understanding resource inside your own resource group and find its Keys and Endpoint page. Copy the endpoint (an address that looks likehttps://your-resource.cognitiveservices.azure.com/) and one key. - In Colab, do not paste the key straight into your code where it is visible. Use Colab’s Secrets panel (the key icon on the left): add
CU_KEYandCU_ENDPOINT, then read them in code. This keeps the secret out of the notebook you might share. - Type this yourself — do not copy-paste. Typing is how your fingers and memory learn:
from google.colab import userdataendpoint = userdata.get('CU_ENDPOINT')key = userdata.get('CU_KEY')
This is the exam-and-employer point in one: a key authorises real spending on your subscription. Keeping it in Secrets instead of in your code is exactly the professional habit that gets you trusted with access.
16:00 – 25:00
Write the four steps
- Import & Connect. Install and import the client, then create it from your endpoint and key. Type it out:
!pip install azure-ai-documentintelligencefrom azure.ai.documentintelligence import DocumentIntelligenceClientfrom azure.core.credentials import AzureKeyCredentialclient = DocumentIntelligenceClient(endpoint, AzureKeyCredential(key))- Send. Point the prebuilt
prebuilt-invoice(orprebuilt-receipt) analyzer at a document by its URL, and wait for the result. poller = client.begin_analyze_document("prebuilt-receipt", {"urlSource": doc_url})result = poller.result()- Use. Loop over the returned documents and print the fields — the loop skill from Week 1:
for d in result.documents:then indentedfor name, field in d.fields.items(): print(name, field.get("content"))- Run it. You just extracted structured fields from a document in code. That is a real, working information-extraction application.
If it errors, read the message slowly — errors are instructions, not insults. A wrong endpoint, a missing key, or a bad document URL are the usual three. Fix one thing at a time.
25:00 – 30:00
Make it yours
- Change one thing and re-run — swap the document URL, or switch
prebuilt-receiptforprebuilt-invoice. Changing something and predicting the result before you run it is how you truly learn. - Add one line that prints only the total, not every field. Small, deliberate edits build real understanding faster than reading ever will.
- Then do the assignment. If you hit a wall, send me the exact error text — do not sit on it silently. Every developer alive gets stuck daily; the good ones just ask sooner.
You have now built a chat client, an agent, a text analyser and an extractor. Read that sentence again. Two days ago you doubted you could. Two days from now you sit the exam.
Key Terms
Four-Step Pattern
Import → Connect → Send → Use. The shape of every AI app in this course, including today’s extractor.
Endpoint & Key
The endpoint is the service’s address; the key is the secret authorising you to call it. Both come from your resource. Never share the key.
Client / SDK
A client object, created from an SDK library, is your code’s connection to the service — the thing you send documents to.
Secrets (Colab)
A safe store for keys, separate from your code, so a shared notebook never leaks your credentials. A basic professional habit.
Today’s Assignment
Build and run a Python program that extracts fields from a document, then explain how it works.
- Part 1 — build it. In Colab, write the four steps to extract fields from a receipt or invoice using a prebuilt analyzer. Store your key in Colab Secrets, not in the code. Send me the Colab link (or clear screenshots of code and output).
- Part 2 — label it. In your code, add a comment above each section marking which of the four steps it is:
# Import,# Connect,# Send,# Use. - Part 3 — explain it. In two or three sentences, tell me why you stored the key in Secrets rather than typing it into the notebook — and what could go wrong if you had not.
- Part 4 — honesty check. Tell me one thing in this build that confused you and how you got past it (or that you are still stuck on). I would rather know.
Submit assignment 23 →
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Self-Check — Exam-Style Questions
1. In the four-step AI pattern, what does the Connect step do?
- A. Prints the final answer to the screen
- B. Creates a client using the endpoint and key so your code can talk to the service
- C. Trains a brand-new model
- D. Deletes the resource group
B. Creates a client using the endpoint and key so your code can talk to the service — Connect builds the client from your endpoint (address) and key (secret). Import brings in the library, Send hands over the document, Use does something with the result.
2. Why should you store your service key in Colab Secrets rather than pasting it directly into a code cell?
- A. It makes the code run faster
- B. It keeps the secret out of a notebook you might share, so it cannot leak
- C. Colab refuses to run without it
- D. It is required to use Python
B. It keeps the secret out of a notebook you might share, so it cannot leak — A visible key in shared code is one of the most common and costly beginner mistakes. Secrets keeps the credential separate from the code, which is exactly the professional habit that earns you trusted access.
3. Your program returns an error. What is the most sensible first move?
- A. Delete everything and start a completely different project
- B. Read the error message carefully — it usually tells you what is wrong
- C. Assume the service is broken and give up
- D. Share your key in a forum to get help
B. Read the error message carefully — it usually tells you what is wrong — Errors are instructions. The three usual causes here are a wrong endpoint, a missing or wrong key, or a bad document URL. Fix one thing at a time. And never, ever paste your key into a public forum (that rules out D).
Exam Objectives Covered
Building extraction appsFoundry / Azure AI SDKEndpoints and keysKey securityPrebuilt analyzersFour-step pattern
Exam Prep · Full ReviewDAY 24
Full Syllabus Review + Microsoft Learn Practice Assessment
Why This Matters
You have learned the whole syllabus. Today you find out what has stuck and, more usefully, what has not — while there is still time to fix it. This is the single most valuable study session of the five weeks, because a practice assessment does not just score you, it hands you a map of your own weak spots. Do not fear a low score today. A gap found on Thursday is a gap closed before the exam. Be honest with yourself in this session. That honesty is what passes exams.
30-Minute Module
0:00 – 8:00
Domain 1 rapid review — AI concepts (40–45%)
- Say the nesting dolls out loud: AI > Machine Learning > Generative AI. If a system learned from examples it is ML; if it creates new content it is generative; if it takes actions with tools it is agentic.
- The five workloads, from memory: generative & agentic, text analysis, speech, computer vision, information extraction. The exam describes a scenario and asks which one — free marks if you know the list.
- The four text-analysis techniques: sentiment analysis, entity detection, key phrase extraction, summarisation. Speech has two directions: recognition (audio→text) and synthesis (text→audio).
- Responsible AI — all six principles. Fairness, Reliability & Safety, Privacy & Security, Inclusiveness, Transparency, Accountability. This is heavily tested. Know a one-line example of each.
If any bullet above made you hesitate, that is a weak spot — write it at the top of a “revise tonight” list right now, before you forget.
8:00 – 15:00
Domain 2 rapid review — Foundry & solutions (55–60%)
- This is the bigger half. Microsoft Foundry is where you find, deploy and use models — the model catalog lists them, a project is your workspace, and a deployment is a model made live for you to call.
- Prompts: a system prompt sets the model’s role and rules; a user prompt is the actual request. Temperature controls randomness — low for factual, higher for creative.
- Tokens are the chunks of text a model reads and writes; embeddings turn text into numbers that capture meaning; inference is the model producing an answer.
- Agents take actions with tools. Content Understanding extracts structured fields from documents, images, audio and video (this week). Everything you build lives inside your own resource group under a Contributor role — a role says what you can do, a scope says where.
Domain 2 is over half the marks and it is the practical half you have actually built. Trust your hands here — you have done this, not just read it.
15:00 – 27:00
Take the official practice assessment
- Go to
learn.microsoft.com, sign in to your own Microsoft Learn profile (the free one from Day 2), open the AI-901 certification page, and start the free Practice Assessment you bookmarked. - Do it properly: no notes, no searching, phone away. This is a rehearsal — treat it like the real thing so today’s score actually means something.
- When you finish, do not just look at the number. Go through every question you got wrong and read the explanation. Write each missed topic on your “revise tonight” list.
- If you run past 30 minutes finishing the assessment, that is fine — this one is worth the extra time. Finish it.
The pass mark is 700 out of 1000 — roughly 70%. If you are near or above that today, you are on track. If you are below, you now know exactly which topics to hit. Either way, today did its job.
27:00 – 30:00
Turn mistakes into a plan
- Look at your “revise tonight” list. Pick your three weakest topics and, for each, find the day in this programme that covered it. You have the material — go back to it tonight.
- Do not try to re-learn everything. Targeted revision of three weak areas beats a panicked reread of all five weeks.
- Then send me your results, per the assignment. Tell me your real score — I am not marking you, I am helping you aim tomorrow’s final drill at what you actually need.
One day to go. You are not cramming new material tomorrow — you are sharpening. That is a good place to be the night before.
Key Terms
Practice Assessment
Microsoft’s free official AI-901 mock exam on Microsoft Learn. Your best gauge of readiness and your clearest map of weak spots.
Six Responsible AI Principles
Fairness; Reliability & Safety; Privacy & Security; Inclusiveness; Transparency; Accountability. Heavily tested — know an example of each.
Pass Mark 700
AI-901 is scored out of 1000; you need 700 — about 70%. It is scaled, not a raw percentage, but 700 is the line.
Targeted Revision
Revising the specific topics you got wrong, not everything. Far more effective than rereading the whole syllabus in a panic.
Today’s Assignment
Take the official practice assessment, then send me your honest results and a plan.
- Part 1 — evidence. Complete the Microsoft Learn AI-901 Practice Assessment under exam conditions. Screenshot your final score.
- Part 2 — diagnosis. List the three topics you scored weakest on, and for each, name the day of this programme that covered it (e.g. “Responsible AI — Days 7 and 8”).
- Part 3 — correction. Pick the single question you got wrong that surprised you most. Write out the correct answer and, in your own words, why it is correct. This is the most valuable thing you will do all week.
- Part 4 — readiness. One honest line: do you feel ready to book the exam, or do you want a few more days? There is no wrong answer — I just need to know.
Submit assignment 24 →
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Self-Check — Exam-Style Questions
1. Which list correctly names the six Responsible AI principles?
- A. Speed, Cost, Accuracy, Scale, Security, Uptime
- B. Fairness; Reliability & Safety; Privacy & Security; Inclusiveness; Transparency; Accountability
- C. Import, Connect, Send, Use, Test, Deploy
- D. Vision, Speech, Text, Audio, Video, Documents
B — Fairness; Reliability & Safety; Privacy & Security; Inclusiveness; Transparency; Accountability. These six come up repeatedly on AI-901. Memorise them and keep a one-line real-world example of each ready.
2. What score do you need to pass exam AI-901?
- A. 500 out of 1000
- B. 600 out of 1000
- C. 700 out of 1000
- D. 900 out of 1000
C. 700 out of 1000 — The pass mark is 700 on a scaled 1000-point score, roughly 70%. It is not a simple raw percentage of questions right, but aiming comfortably above 700 on your practice assessment is the right target.
3. You score 62% on the practice assessment. What is the best use of your remaining study time?
- A. Reread all five weeks from the start in a panic
- B. Give up — you cannot pass
- C. Identify your weakest topics and revise those specifically
- D. Take the real exam immediately without any more study
C. Identify your weakest topics and revise those specifically — A below-pass practice score is not a verdict, it is a map. Targeted revision of the specific areas you missed is far more effective than rereading everything or rushing into the real thing underprepared.
Exam Objectives Covered
Full Domain 1 reviewFull Domain 2 reviewResponsible AIPractice assessmentExam scoringTargeted revision
Exam Prep · ReadinessDAY 25
Exam Day Readiness: Final Drill, Booking, and What Comes Next
Why This Matters
Day 25. You got here — twenty-five days, thirty minutes at a time. Understand what that proves: you can learn hard things by showing up daily. That habit is worth more than any single certificate. Today we do a final drill, I show you exactly how to book and sit the exam, and we talk about what comes after — because AI-901 is not the finish line, it is the first rung. I am proud of you. Now let us go and get this done.
30-Minute Module
0:00 – 8:00
Final drill — the highest-yield facts
- Run these fast, out loud. Nesting: AI > ML > Generative; agentic = takes actions. Five workloads: generative & agentic, text analysis, speech, computer vision, information extraction.
- Text analysis four: sentiment, entity detection, key phrases, summarisation. Speech two: recognition (audio→text), synthesis (text→audio). Vision: interpreting images; image generation creates them.
- Responsible AI six: Fairness; Reliability & Safety; Privacy & Security; Inclusiveness; Transparency; Accountability. RBAC: role = what, scope = where.
- Foundry: catalog (find models), project (workspace), deployment (model made live). Prompts: system sets rules, user makes the request; temperature = randomness. Tokens, embeddings, inference. Content Understanding: extracts fields from documents, images, audio, video.
This bullet block is your one-page cheat sheet. Read it this morning, and once more an hour before your exam. Do not learn anything new today.
8:00 – 16:00
How to book the exam — step by step
- Go to
learn.microsoft.com, sign in to your own Microsoft Learn profile, and open the AI-901 certification page. Click Schedule exam. Microsoft books through Pearson VUE, its testing partner. - Choose online proctored so you can sit it from home — a real person supervises you over your webcam. (An in-person test centre is the alternative if you prefer.) The current price is around USD 99; check the exact fee and any local pricing on the page before you pay.
- What you need: a quiet, private room; a laptop with a working webcam and microphone; a stable internet connection; and a valid photo ID whose name matches your profile. You will run a quick system check beforehand — do it a day early, not five minutes before.
- The exam is roughly 40–60 questions in about 45 minutes, multiple choice and similar formats. You get your result on screen immediately. Pass, and the certificate lands in your Microsoft Learn profile, in your name, verifiable by any employer — forever.
Book it for a specific day and tell me the date. A booked exam is a commitment; an “I’ll do it soon” drifts forever. Pick the day.
16:00 – 25:00
On the day — tactics that win marks
- Read every question twice. AI-901 questions are fair but precise — words like “NOT”, “best”, and “least” change everything. The near-neighbour traps (recognition vs synthesis, prebuilt vs custom, role vs scope) are exactly the ones we drilled.
- Never leave a blank. There is no penalty for a wrong answer, so answer everything. If unsure, eliminate the obviously wrong options first, then choose the best of what remains.
- Flag and move on. If a question is hard, mark it for review and keep going — do not burn five minutes on one mark while ten easy ones wait. Come back at the end.
- Trust your build weeks. Over half the exam is Domain 2, the practical half you actually did with your own hands. When a scenario question appears, picture the playground or your Colab notebook — you have been there.
Nerves are normal and they are not a problem — slow your breathing, read carefully, and let your preparation do the work. You did not cram this; you built it over five weeks.
25:00 – 30:00
What comes next
- The moment you pass, add “Microsoft Certified: Azure AI Fundamentals (AI-901)” to your CV and LinkedIn, with the verification link from your Learn profile. That credential is now a fact about you, not a hope.
- Your next rung: I recommend AZ-900 (Azure Fundamentals) next — it broadens your cloud foundation and pairs naturally with AI-901 on a CV. After that, AI-102 (Azure AI Engineer Associate) is the deeper, more hireable step that builds directly on everything you did these five weeks.
- Keep the daily habit. Thirty minutes a day is how you got here; it is how you will get to the next one too. Do not let the streak die with the exam.
- And whatever your result on the day — email me. If you pass, we celebrate and plan AZ-900. If you do not pass first time, we look at the score report together and go again. Nobody who keeps showing up stays down.
You started this with zero technical background. In twenty-five days you learned the cloud, Python, and how to build with AI. Remember that the next time something looks too hard.
Key Terms
Pearson VUE
Microsoft’s testing partner. You book and sit AI-901 through it, online-proctored from home or at a test centre.
Online Proctored
Sitting the exam from home while a supervisor watches via webcam. Needs a quiet room, working camera and mic, and photo ID.
Immediate Result
AI-901 shows your pass/fail on screen when you finish. A pass places the certificate in your Learn profile, in your name.
Next Steps
AZ-900 to broaden your cloud foundation, then AI-102 to go deeper into building AI. AI-901 is the first rung, not the top.
Today’s Assignment
Book your exam — or set the date — and send me your readiness confirmation and your plan for after.
- Part 1 — commit. Go to the AI-901 page on Microsoft Learn and either book your exam or choose a firm date to book it by. Tell me the date. If anything blocks the booking (ID, payment, system check), tell me exactly what — that is mine to help with.
- Part 2 — the cheat sheet. Write out, from memory, the five workloads and the six Responsible AI principles. If you can do this cleanly, you are ready.
- Part 3 — the trap. Write one sentence each distinguishing: speech recognition vs synthesis, prebuilt vs custom analyzer, and role vs scope. These are the exam’s favourite traps — own them.
- Part 4 — forward. One paragraph: what job or opportunity you want this certificate to open, and which certification you plan to take next.
Submit assignment 25 →
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Your work is marked against the published rubric and comes back to you as a written letter, usually within minutes. It is also saved to your progress page, so you can re-read every letter later. Do not submit until you have passed the self-check below.
Self-Check — Exam-Style Questions
1. Through which partner do you book and sit exam AI-901?
- A. Amazon AWS
- B. Pearson VUE
- C. Google Cloud
- D. Microsoft Foundry
B. Pearson VUE — Microsoft schedules its certification exams through Pearson VUE. You start from the AI-901 page on Microsoft Learn and can choose online-proctored (from home) or an in-person test centre.
2. You are unsure of an answer during the exam. There is no penalty for a wrong answer. What should you do?
- A. Leave it blank to be safe
- B. Eliminate the wrong options and choose the best remaining answer — never leave it blank
- C. End the exam early
- D. Answer every question with option A
B. Eliminate the wrong options and choose the best remaining answer — never leave it blank — With no penalty for wrong answers, a blank can only score zero while a considered guess might score. Eliminate the clearly wrong choices first to improve your odds, and flag hard questions to revisit at the end.
3. You pass AI-901. Which certification is the natural deeper next step that builds directly on it?
- A. AI-102 (Azure AI Engineer Associate)
- B. A qualification in a completely unrelated field
- C. There is nothing beyond AI-901
- D. You must retake AI-901 every month
A. AI-102 (Azure AI Engineer Associate) — AI-102 is the deeper, more hireable step that builds on the AI concepts and Foundry skills you learned here. Broadening first with AZ-900 (Azure Fundamentals) is also a strong move. AI-901 is the first rung, not the top of the ladder.
Exam Objectives Covered
Full syllabus consolidationExam booking (Pearson VUE)Exam-day strategyResponsible AI recallWorkload recallCertification pathway