AI-900 in a Weekend: A Total Beginner's No-BS Guide to Microsoft's AI Fundamentals Cert
I couldn't explain the difference between machine learning and deep learning two weekends ago. Now I have an AI certification. Here's exactly how that happened.
Why the AI-900 Exists
Microsoft made this exam for people who aren't data scientists but need to understand AI. Think project managers, business analysts, IT admins — anyone who hears "we're implementing AI" at work and nods while internally panicking.
The AI-900 gives you enough knowledge to hold an intelligent conversation about AI without pretending to know how to train a neural network.
What's Actually on the Exam
Four buckets:
AI workloads and considerations (15-20%). What AI can and can't do. Responsible AI principles. Microsoft asks surprisingly philosophical questions here — like identifying bias in training data.
Machine learning fundamentals (20-25%). Regression vs. classification vs. clustering. You don't need to code anything. You need to know that regression predicts numbers, classification predicts categories, and clustering groups similar things. That's genuinely 80% of it.
Computer vision (15-20%). What Azure's Computer Vision, Custom Vision, and Face API do. OCR. Object detection vs. image classification.
NLP and generative AI (25-30%). This section got beefed up recently. Azure OpenAI Service, language understanding, speech services. They're testing whether you know what these services DO, not how to build them.
The Weekend Speedrun
Saturday morning: Microsoft Learn's AI-900 learning path. It's 5 modules, maybe 6 hours total. Made coffee. Read everything. Took notes on Azure-specific service names.
Saturday afternoon: Free AI-900 practice questions on ExamCert. First attempt: 72%. Not great. Identified my weak spots — I kept confusing Azure Cognitive Services (old name) with Azure AI Services (new name).
Sunday morning: Reviewed weak areas. Watched a couple YouTube videos on responsible AI (Microsoft's favorite topic). Did another round of practice questions. Hit 88%.
Sunday afternoon: Sat the exam online from my couch. Passed with 847/1000.
Total study time: maybe 12 hours. Total cost: $165 exam + $4.99 for ExamCert practice tests. That's it.
Three Mistakes to Avoid
1. Don't study deep learning math. You don't need to understand backpropagation. The exam asks "what IS deep learning?" not "how does it work mathematically." Know that it uses neural networks with multiple layers. Done.
2. Know the NEW service names. Microsoft rebranded everything. "Cognitive Services" → "Azure AI Services." "Form Recognizer" → "Azure AI Document Intelligence." The exam uses the new names. Old study materials will confuse you.
3. Don't skip responsible AI. I know it sounds fluffy. But 2-3 questions will directly ask about fairness, transparency, accountability, and inclusiveness. Free points if you studied. Lost points if you didn't.
Who This Is Actually For
Career changers who want "AI" on their resume without spending six months
IT pros who manage Azure and want to understand the AI services menu
Students building a certification portfolio
Anyone whose boss said "learn about AI" — this is your receipt
Who Should Skip It
Anyone with a DP-100 or AI-102 already — you're past this
People who want hands-on ML engineering skills — this cert won't teach you that
The Cost Argument
$165. No prerequisites. No required training. Microsoft Learn is free. Practice tests on ExamCert are $4.99 with a money-back guarantee.
You could literally go from "zero AI knowledge" to "Microsoft AI certified" in one weekend for under $170.
So You Want to Pass the AI-900? Here's What Actually Matters
I passed the Azure AI Fundamentals (AI-900) exam a while back, and honestly, it's one of the more approachable Microsoft certs out there. If you're curious about AI but don't have a deep technical background, this is a solid starting point. Here's the quick rundown.
What is the AI-900?
It's Microsoft's entry-level certification for Azure AI services. You don't need to be a data scientist or a developer to pass it. The exam tests whether you understand core AI concepts, how machine learning works at a high level, and what Azure offers for things like computer vision, natural language processing, and generative AI.
It costs $99 USD and you get about 45 minutes to answer roughly 33 questions. Passing score is 700 out of 1000.
The Five Domains You Need to KnowThe exam breaks down into five areas, and the weight distribution matters more than people think:
1. AI workloads and considerations (15-20%) - ethical AI principles, responsible AI, common AI scenarios
2. Machine learning fundamentals on Azure (20-25%) - this is the biggest chunk, so spend extra time here
3. Computer vision on Azure (15-20%) - image classification, object detection, Azure AI Vision
4. NLP on Azure (15-20%) - text analytics, language understanding, speech services5. Generative AI on Azure (15-20%) - Azure OpenAI Service, prompt engineering basics, copilots
The ML section carries the most weight. I'd say about 40% of my study time went there, and it paid off.
Three Study Tips That Actually WorkFirst, go through the free Microsoft Learn path for AI-900. Seriously, a huge chunk of exam questions pull directly from that material. It's not the most exciting read, but it's accurate and complete.
Second, watch the FreeCodeCamp AI-900 video on YouTube. It's a few hours long and covers everything in a way that's easier to absorb than reading docs. I played it at 1.5x speed during commutes.
Third, do practice questions before you sit the exam. This is where most people mess up - they study the concepts but don't practice the actual question format. Microsoft loves scenario-based questions where you pick the right Azure service for a given problem.
Practice exams at https://www.examcert.app/exams/azure-ai-900/ are great for getting used to that style.
Best Resources
- Microsoft Learn AI-900 learning path (free, official)
- FreeCodeCamp YouTube course (free, great explanations)
- ExamCert practice tests at https://www.examcert.app/exams/azure-ai-900/ (exam-style questions)
Bottom Line
Most people can prep for this in 1-2 weeks of casual study. It's not a hard exam if you put in the time. But don't skip the practice questions - knowing concepts is different from knowing how Microsoft phrases things.
Book your exam, give yourself a deadline, and go get certified.
在今快速發展的科技時代，人工智慧 (AI) 已成為各行各業不可或缺的技術。對於希望進入 AI 領域，或證明其在 Microsoft Azure AI 基礎知識方面能力的人士來說，Microsoft Azure AI-900 認證考試是一項重要的入門級證照。
這項認證考試主要旨在評估考生對 Azure AI 服務基礎概念的理解。它涵蓋了 AI 的核心工作負載，包括：
• 機器學習 (Machine Learning)：這是 AI 系統的基礎，專注於「指導」電腦模型進行預測並從資料中得出結論。考試會涉及監督式學習（如分類和迴歸）、非監督式學習（如聚類）等概念。
• 電腦視覺 (Computer Vision)：這部分探討 AI 系統如何透過相機、影片和影像來視覺化解讀世界，常見應用包括影像分類、物件偵測、語意分割、影像分析以及臉部偵測與辨識。
• 自然語言處理 (Natural…
Unlock the Power of AI: Your Journey Starts with the AI-900 Certification
Thinking about dipping your toes into the exciting world of Artificial Intelligence (AI)? The Microsoft Azure AI Fundamentals certification (AI-900) is your perfect launchpad! Gain foundational knowledge of AI concepts & Azure AI services. This bite-sized cert empowers you to contribute to AI projects & unlock a world of possibilities.
It supports a maximum of 8640 data points. Break this down into smaller requests to improve the performance.
When to use Anomaly Detector
Process the algorithm against an entire set of data at one time
It creates a model based on your complete data set and the finds anomalies
Uses streaming data by comparing previously seen dat points to the last datapoint to determine if your latest one is an anomaly.
Model is created using the data points you send and determines if the current point is an anomaly.
Microsoft Azure AI Fundamentals: Explore natural language processing
Analyze Text with the Language Service
Used to describe solutions that involve extracting information from large volumes of unstructured data.
Analyzing text is a process to evaluate different aspects of a document or phrase, to gain insights about that text.
Text Analytics Techniques
Interpret words like “power”, “powered”, and “powerful” as the same word.
Convert to tree like structures (Noun phrases)
Often used for sentiment analysis
Determine the language of a document or text
Perform sentiment analysis (positive or negative)
Extract key phrases from text to indicate key talking points
Identify and categorize entities (places, people, organizations, etc)
Get started with Text analysis
Language name
ISO 6391 language code
Score as a level of confidence n the language returned.
Evaluates text to return a sentiment score and labels for each sentence
Useful for detecting positive or negative sentiment
Classification is between 0 to 1 with 1 being most positive
A score of 0.5 is indeterminate sentiment.
The phrase doesn’t have sufficient information to determine the sentiment.
Mixing language content with the language you tell it will return 0.5 also
Key Phrase extraction
Used to determine the main talking points of a text or a document
Depending on the volume this can take longer, so you can use the key phrase extraction capabilities of the Language Service to summarize main points.
Key phrase extraction can provide context about the document or text
Entity Recognition
Person
Location
OrganizationQuantity
DateTime
URL
Email
US-based phone number
IP address
Recognize and Synthesize Speech
Acoustic model - converts audio signal to phonemes (representation of specific sounds)
Language model - maps the phonemes to words using a statistical algorithm to predict the most probably sequence of words based on the phonemes
ability to generate spoken output
Usually converting text to speech
This process tokenizes the set to break it down into individual words, assign phonetic sounds to each word
It then breaks the phonetic transcription to prosodic units to create phonemes for the audio
Get started with speech on Azure
Use this for demos, presentations, or scenarios where a person is speaking
In real time it can translate to many lunges as it processes
Audio files with Shared access signature (SAS) URI can be used and results are received asynchronously.
Jobs will start executing within minutes, but no estimate is provided for when the job changes to running state
Used to convert text to speech
Voices can be selected that will vocalize the text
Custom voices can be developed
Voices are trained using neural networks to overcome limitations in speech synthesis with regards to intonation.
Translate Text and Speech
Where each word is translated to the corresponding word in the target language
This approach has issues. For example, a direct word to word translation may not exist or the literal translation may not be the correct meaning of the phrase
Machine learning has to also understand the semantic context of the translation.
This provides more accurate translation of the input phrase or phrases
Grammar, formal versus informal, colloquialism all need to be considered
Text and speech translation
Profanity filtering - remove or do not translate profamity
Selective translation - tag content that isn’t to be translated (brand names, code names, etc)
Speech to text - transcribe speech from an audio source to text format.
Text to speech - used to generate spoken audio from a text source
Speech translation - translate speech in one language to text or speech in another
Create a language model with Conversational language Understanding
A None intent exists.
This should be used when no intent has been identified and should provide a message to a user.
Getting started with Conversational Language Understanding
Authoring the model - Defining entities, intents, and utterances to use to train the model
Entity Prediction - using the model after it is published.
Define intents based on actions a user would want to perform
Each intent should include a variety of utterances as examples of how a user may express the intent
If the intent can be applied to multiple entities, include sample utterances for each potential entity.
Machine-Learned - learned by the model during training from context in the sample utterances you provide
List - Defined as a hierarchy of lists and sublists
RegEx - regular expression patterns
Pattern.any - entities used with patterns to define complex entities that may be hard to extract from sample utterances
After intents and entities are created you train the model.
Training is the process of using your sample utterances to teach the model to match natural language expressions that a user may say to probable intents and entities.
Training and testing are iterative processes
If the model does not match correctly, you create more utterances, retrain, and test.
When results are satisfactory, you can publish the model.
Client applications can use the model by using and endpoint for the prediction resource
Build a bot with the Language Service and Azure Bot Service
Knowledge base of question and answer pairs. Usually some built-in natural language processing model to enable questions and can understand the semantic meaning
Bot service - to provide an interface to the knowledge base through one or more channels
Microsoft Azure AI Fundamentals: Explore knowledge mining
Used to describe solutions that involve extracting information from large volumes of unstructured data.
It has a services in Cognitive services to create a user-managed index.
The index can b meant for internal use only or shared with the public.
It can use other Cognitive Services capabilities to extract the information
What is Azure Cognitive Search?
Provides a programmable search engine build on Apache Lucene
Highly available platform with 99.9% uptime SLA for cloud and on-premise assets
Data from any source - accepts data form any source provided in JSON format with auto crawling support for selected data sources in Azure
Full text search and analysis - Offers full text search capabilities supporting both simple query and full Lucene query syntax
AI Powered search - has Cognitive AI capabilities built in for image and text analysis from raw content
Multi-lingual - offers linguistic analysis for 56 langues
Geo-enabled - supports geo-search filtered based on proximity to a physical location
Configurable user experience - it includes capabilities to improve the user experience (autocomplete, autosuggest, pagination, hit highlighting, etc)
Identify elements of a search solution
Folders with files,
Text in a database
Etc
Use a skillset to Define an enrichment pipeline
Key Phrase Extraction - uses a pre-trained model to detect important phrases based on term placement, linguistic rules, proximity to terms
Text Translation - pre-trained model to translate the input text into various languages for normalization or localization use cases
Image Analysis Skills - uses an image detection algorithm to identify the content of an image an generate a text description
Optical Character Recognition Skills - extract printed or handwritten text from images, photos, videos
Understand indexes
Index schema - index includes a definition of the structure of the data in the documents to read.
Index attributes - Each field in a document the index stores its name, the data type, supported behaviors (searchable, sortable, etc)
Best indexes use only the features that are required/needed
Use an indexer to build an index
Push method - JSON data is pushed into a search index via a REST API or a .NET SDK. Most flexible and with least restrictions
Pull method - Search service indexer pulls from popular Azure data sources and if necessary exports the Tinto JSON if its not already in that format
Use the pull method to load data with an indexer
Azure Cognitive search’s indexer is a crawler that extracts searchable text and metadata form an external Azure data source an populates a search index using field-to-field mapping between the data and the index.
Data import monitoring and verification
Indexers only import new or updated documents. It is normal to see zero documents indexed
Health information is displayed in a dashboard.
You can monitor the progress of the indexing
Making changes to an index
You need to drop and recreate indexes if you need to make changes to the field definitions
An approach to update your index without impacting your users is to create a new index with a new name
After importing data, switch to the new index.
Persist enriched data in a knowledge store
A knowledge store is persistent storage of enriched content.
The knowledge store is to store the data generated from Ai enrichment in a container.
Microsoft Azure AI Fundamentals: Explore visual studio tools for machine learning
What is machine learning? A technique that uses math and statistics to create models that predict unknown values
Types of Machine learning
Regression - predict a continuous value, like a price, a sales total, a measure, etc
Classification - determine a class label.
Clustering - determine labels by grouping similar information into label groups
x = features
y = label
Azure Machine Learning Studio
You can use the workspace to develop solutions with the Azure ML service on the web portal or with developer tools
Web portal for ML solutions in Sure
Capabilities for preparing data, training models, publishing and monitoring a service.
First step assign a workspace to a studio.
Compute targets are cloud-based resources which can run model training and data exploration processes
Compute Instances - Development workstations that data scientists can use to work with data and models
Compute Clusters - Scalable clusters of VMs for on demand processing of experiment code
Inference Clusters - Deployment targets for predictive services that use your trained models
Attached Compute - Links to existing Azure compute resources like VMs or Azure data brick clusters
What is Azure Automated Machine Learning
Jobs have multiple settings
Provide information needed to specify your training scripts, compute target and Azure ML environment and run a training job
Understand the AutoML Process
ML model must be trained with existing data
Data scientists spend lots of time pre-processing and selecting data
This is time consuming and often makes inefficient use of expensive compute hardware
In Azure ML data for model training and other operations are encapsulated in a data set.
You create your own dataset.
Classification (predicting categories or classes)
Regression (predicting numeric values)
Time series forecasting (predicting numeric values at a future point in time)
After part of the data is used to train a model, then the rest of the data is used to iteratively test or cross validate the model
The metric is calculated by comparing the actual known label or value with the predicted one
Difference between the actual known and predicted is known as residuals; they indicate amount of error in the model.
Root Mean Squared Error (RMSE) is a performance metric. The smaller the value, the more accurate the model’s prediction is
Normalized root mean squared error (NRMSE) standardizes the metric to be used between models which have different scales.
Shows the frequency of residual value ranges.
Residuals represents variance between predicted and true values that can’t be explained by the model, errors
Most frequently occurring residual values (errors) should be clustered around zero.
You want small errors with fewer errors at the extreme ends of the sale
Should show a diagonal trend where the predicted value correlates closely with the true value
Dotted line shows a perfect model’s performance
The closer to the line of your model’s average predicted value to the dotted, the better.
Services can be deployed as an Azure Container Instance (ACI) or to a Azure Kubernetes Service (AKS) cluster
For production AKS is recommended.
Identify regression machine learning scenarios
Regression is a form of ML
Understands the relationships between variables to predict a desired outcome
Predicts a numeric label or outcome base on variables (features)
Regression is an example of supervised ML
What is Azure Machine Learning designer
Allow you to organize, manage, and reuse complex ML workflows across projects and users
Pipelines start with the dataset you want to use to train the model
Each time you run a pipelines, the context(history) is stored as a pipeline job
Encapsulates one step in a machine learning pipeline.
Like a function in programming
In a pipeline project, you access data assets and components from the Asset Library tab
You can create data assets on the data tab from local files, web files, open at a sets, and a datastore
Data assets appear in the Asset Library
Azure ML job executes a task against a specified compute target.
Jobs allow systematic tracking of your ML experiments and workflows.
Understand steps for regression
To train a regression model, your data set needs to include historic features and known label values.
Use the designer’s Score Model component to generate the predicted class label value
Connect all the components that will run in the experiment
Average difference between predicted and true values
It is based on the same unit as the label
The lower the value is the better the model is predicting
The square root of the mean squared difference between predicted and true values
Metric based on the same unit as the label.
A larger difference indicates greater variance in the individual label errors
Relative metric between 0 and 1 on the square based on the square of the differences between predicted and true values
Closer to 0 means the better the model is performing.
Since the value is relative, it can compare different models with different label units
Relative metric between 0 and 1 on the square based on the absolute of the differences between predicted and true values
Closer to 0 means the better the model is performing.
Can be used to compare models where the labels are in different units
Also known as R-squared
Summarizes how much variance exists between predicted and true values
Closer to 1 means the model is performing better
Remove training components form your data and replace it with a web service inputs and outputs to handle the web requests
It does the same data transformations as the first pipeline for new data
It then uses trained model to infer/predict label values based on the features.
Create a classification model with Azure ML designer
Classification is a form of ML used to predict which category an item belongs to
Like regression this is a supervised ML technique.
Understand steps for classification
True Positive - Model predicts the label and the label is correct
False Positive - Model predicts wrong label and the data has the label
False Negative - Model predicts the wrong label, and the data does have the label
True Negative - Model predicts the label correctly and the data has the label
For multi-class classification, same approach is used. A model with 3 possible results would have a 3x3 matrix.
Diagonal lien of cells were the predicted and actual labels match
Number of cases classified as positive that are actually positive
True positives divided by (true positives + false positives)
Fraction of positive cases correctly identified
Number of true positives divided by (true positives + false negatives)
Overall metric that essentially combines precision and recall
Classification models predict probability for each possible class
For binary classification models, the probability is between 0 and 1
Setting the threshold can define when a value is interpreted as 0 or 1. If its set to 0.5 then 0.5-1.0 is 1 and 0.0-0.4 is 0
Recall also known as True Positive Rate
Has a corresponding False Positive Rate
Plotting these two metrics on a graph for all values between 0 and 1 provides information.
Receiver Operating Characteristic (ROC) is the curve.
In a perfect model, this curve would be high to the top left
Area under the curve (AUC).
Remove training components form your data and replace it with a web service inputs and outputs to handle the web requests
It does the same data transformations as the first pipeline for new data
It then uses trained model to infer/predict label values based on the features.
Create a Clustering model with Azure ML designer
Clustering is used to group similar objects together based on features.
Clustering is an example of unsupervised learning, you train a model to just separate items based on their features.
Understanding steps for clustering
Prebuilt components exist that allow you to clean the data, normalize it, join tables and more
Requires a dataset that includes multiple observations of the items you want to cluster
Requires numeric features that can be used to determine similarities between individual cases
Initializing K coordinates as randomly selected points called centroids in an n-dimensional space (n is the number of dimensions in the feature vectors)
Plotting feature vectors as points in the same space and assigns a value how close they are to the closes centroid
Moving the centroids to the middle points allocated to it (mean distance)
Reassigning to the closes centroids after the move
Repeating the last two steps until tone.
Maximum distances between each point and the centroid of that point’s cluster.
If the value is high it can mean that cluster is widely dispersed.
With the Average Distance to Closer Center, we can determine how spread out the cluster is
Remove training components form your data and replace it with a web service inputs and outputs to handle the web requests
It does the same data transformations as the first pipeline for new data
It then uses trained model to infer/predict label values based on the features.
