My AI Journey Begins: Passing AI-900 Azure AI Fundamentals

Part 1 of the "From .NET to AI Engineer" series

The Wake-Up Call

It's been almost a year since I passed AI-900, and looking back, it was the decision that kickstarted my entire AI journey. Let me take you back to early 2025.

I was building Glucoplate and experimenting with multiple AI services for receipt scanning—Azure OCR, Amazon Textract, even Claude Haiku for text extraction. For meal image analysis, I'd tested OpenAI's vision models, Google Gemini, and Azure AI Vision. I was swimming in AI APIs, but realized something uncomfortable: I was doing AI work without really understanding AI.

Sure, I could call an API. I could compare results. But when I needed to decide why Gemini worked better than GPT-4 Vision for our food recognition use case, I was guessing. When someone asked me to explain the difference between classification and regression, I stumbled.

That's when I decided to go back to basics.

Why AI-900?

The AI-900: Azure AI Fundamentals certification seemed like the perfect starting point. Microsoft positions it as an entry-level exam that doesn't require coding experience, but don't let that fool you—it's packed with concepts that even experienced developers might not fully grasp.

Here's what convinced me:

Foundation for AI-102: I knew I eventually wanted the Azure AI Engineer Associate certification. AI-900 isn't a prerequisite, but it builds the vocabulary and mental models you'll need.
$99 and a few weeks of study: Low commitment, high value. The certification never expires once you pass.
Credibility in conversations: When discussing AI solutions with stakeholders, having formal certification carries weight.

What AI-900 Actually Covers

The exam breaks down into four main areas:

1. Describe Artificial Intelligence Workloads and Considerations (15-20%)

This is where you learn the "why" of AI. What problems does AI solve? When should you use machine learning vs. rule-based systems? And critically—responsible AI principles.

Microsoft hammers home their Responsible AI principles:

Fairness
Reliability & Safety
Privacy & Security
Inclusiveness
Transparency
Accountability

Why this matters: Every serious AI conversation I've had since—whether with colleagues, in meetings, or when evaluating vendors—touches on responsible AI. It's not just ethics theater—it's a genuine concern as AI systems make real decisions about real people.

2. Describe Fundamental Principles of Machine Learning (20-25%)

This section demystified concepts I'd been hand-waving for years:

Regression vs. Classification vs. Clustering

Regression: Predict a NUMBER
  "What will the temperature be tomorrow?" → 72.5°F

Classification: Predict a CATEGORY
  "Is this email spam?" → Yes/No
  "What type of flower is this?" → Rose/Tulip/Daisy

Clustering: Find GROUPS (no labels!)
  "Group these customers by behavior" → Cluster A, B, C

Supervised vs. Unsupervised Learning

Supervised: You have labeled data. "Here are 1000 emails marked spam/not-spam, learn the pattern."
Unsupervised: No labels. "Here are customer transactions, find interesting patterns."

Key Azure Service: Azure Machine Learning is the platform for building, training, and deploying ML models. You don't need to code for AI-900, but understanding the workflow matters.

3. Describe Features of Computer Vision Workloads (15-20%)

Azure AI Vision is more capable than I realized:

Image Classification: "This is a photo of a cat"
Object Detection: "There's a cat at coordinates (x, y) with bounding box"
Optical Character Recognition (OCR): Extract text from images
Face Detection: Detect faces (note: Face Recognition has been restricted due to ethical concerns)

Real-world connection: Building Glucoplate's meal recognition, I tested Azure AI Vision, OpenAI GPT-4 Vision, and Google Gemini side-by-side. Gemini won for our use case—better food identification accuracy and more reasonable pricing. But understanding why required knowing what these models actually do under the hood.

4. Describe Features of Natural Language Processing (15-20%)

NLP is where Azure AI really shines for enterprise developers:

Sentiment Analysis: Is this review positive or negative?
Key Phrase Extraction: What are the main topics?
Named Entity Recognition (NER): Find people, places, organizations in text
Language Detection: What language is this text?
Translation: Azure Translator supports 100+ languages

Service to know: Azure AI Language (formerly Text Analytics) bundles most NLP capabilities.

5. Describe Features of Generative AI Workloads (20-25%)

This section was updated significantly in 2025 to reflect the GenAI boom:

Azure OpenAI Service: Access GPT models through Azure with enterprise security
Prompt Engineering Basics: Zero-shot, few-shot, chain-of-thought
Responsible GenAI: Content filtering, grounding, and hallucination mitigation

Key concept: Grounding—connecting AI responses to actual data sources—is critical. An ungrounded GPT model will confidently make things up.

My Study Approach

Total time: 3 weeks, about 1-2 hours per day

Resources I used:

Microsoft Learn Paths (Free): The official learning path is comprehensive. I completed every module.
Practice Tests: I used the practice assessment on Microsoft Learn. The real exam felt very similar.
Hands-on Labs: Even though AI-900 doesn't require coding, I spun up Azure AI services and played with them. Seeing sentiment analysis work on real text cemented the concepts.

Exam Day

The exam was 45 minutes with about 40 questions. Format included:

Multiple choice
Drag and drop (match service to use case)
Yes/No scenario questions

I finished in about 30 minutes and passed with a comfortable margin. The questions were fair—if you've done the Microsoft Learn modules, you'll recognize the scenarios.

What I Learned Beyond the Cert

The real value wasn't the certificate (though that's nice). It was:

Vocabulary: I can now have informed conversations about ML approaches without faking it.
Service mapping: I know which Azure service to reach for. Text analysis? Azure AI Language. Image recognition? Azure AI Vision. Full ML pipeline? Azure Machine Learning.
Responsible AI awareness: I now think about fairness and bias in every AI feature I build.

What's Next

AI-900 gave me the foundation. Now I'm preparing for AI-102: Azure AI Engineer Associate—that's where the real engineering happens.

In the next post, I'll share my AI-102 study plan and dive deep into what it takes to actually build AI solutions on Azure.

Key Takeaways

AI-900 is an excellent starting point for any developer entering the AI space
The exam covers AI fundamentals, ML basics, computer vision, NLP, and generative AI
Microsoft's Responsible AI principles matter in real-world AI discussions—learn them
The $99 investment and ~3 weeks of study time is worth it for the foundation it provides

Have questions about AI-900 or your own AI learning journey? Connect with me on Twitter.