Study Strategies and Exam Tips for the AI-102
Key Takeaways
- Weight your study time toward Plan and Manage (20-25%) plus the three 15-20% domains; Agentic (5-10%) is small but easy points.
- AI-102 is a hands-on exam — practice provisioning resources, calling REST/SDK endpoints, and reading code, not just memorizing.
- Know exact service boundaries: Image Analysis vs Custom Vision, Language vs Azure OpenAI, AI Search vs Document Intelligence.
- Budget about 2 minutes per standalone question and answer all linked case-study items before leaving that case study.
- Use the free Microsoft Learn practice assessment and the official GitHub labs; both track the current Foundry-era blueprint.
Last updated: June 2026
Build a Weighted Study Plan
The fastest path to a pass is allocating hours in proportion to the current Skills Measured weights, then layering in lab time. Note that this blueprint differs from older guides: Plan and Manage is now the largest area, and there is a new Agentic domain.
| Skill Area | Weight | Suggested Hours (6-week plan) |
|---|---|---|
| Plan and manage an Azure AI solution | 20-25% | 20-26 |
| Implement generative AI solutions | 15-20% | 16-20 |
| Implement an agentic solution | 5-10% | 6-9 |
| Implement computer vision solutions | 10-15% | 10-14 |
| Implement natural language processing solutions | 15-20% | 16-20 |
| Implement knowledge mining and information extraction | 15-20% | 16-20 |
| Practice exams and review | — | 14-18 |
| Total | 100% | ~98-127 |
Week-by-Week Schedule
Week 1 — Plan and Manage (the new heavyweight)
- Service selection: which Foundry service fits a vision, NLP, speech, info-extraction, or generative task.
- Provisioning a multi-service vs single-service AI resource; finding the default endpoint and keys.
- Security: prefer managed identity / Microsoft Entra auth over account keys; rotate and store keys in Key Vault.
- Responsible AI: Content Safety, content filters, blocklists, and prompt shields.
Week 2 — Generative AI with Azure OpenAI and Foundry
- Provision an Azure OpenAI resource, deploy a model (for example GPT-4o), and call the chat completions API.
- Control output with
temperature,top_p,max_tokens, and stop sequences. - Build a RAG pattern grounding a model in your own data; use prompt templates and prompt flow.
Week 3 — Agentic Solutions (small domain, high ROI)
- Understand when an agent beats a single prompt; configure resources for the Foundry Agent Service.
- Build a single agent, then a multi-agent workflow with orchestration; test and deploy it.
Week 4 — Computer Vision
- Image Analysis visual features (tags, objects, captions, OCR/Read) versus Custom Vision training.
- Choose classification vs object detection; label, train, evaluate, publish, and consume a custom model.
- Video Indexer insights and Spatial Analysis for people movement.
Week 5 — NLP and Knowledge Mining
- Language service: key phrases, entities, sentiment, language detection, PII, and Translator.
- Speech: speech-to-text, text-to-speech, SSML, custom speech, and speech translation.
- AI Search: data source -> indexer -> skillset -> index; semantic and vector search; Knowledge Store projections.
- Document Intelligence prebuilt + custom + composed models; Content Understanding pipelines.
Week 6 — Review and Timed Practice
- Take the free Microsoft Learn practice assessment and 2-3 timed full-length tests; review every miss against the docs.
Free Study Resources
| Resource | What It Gives You | Cost |
|---|---|---|
| Microsoft Learn AI-102 learning paths | Official modules mapped to current objectives | Free |
| Microsoft Learn practice assessment | Sample items in the real question styles | Free |
| Azure free account | 12 months of free services + $200 credit (30 days) | Free |
| Official GitHub AI-102 lab exercises | Guided hands-on labs per skill | Free |
| This study guide | Full AI-102 coverage with practice questions | Free |
Hands-On Practice Is Non-Negotiable
AI-102 tests doing, not reciting. Before exam day you should have personally:
- Provisioned an AI services resource via portal and Azure CLI.
- Called a Vision, Language, and OpenAI endpoint over REST with correct headers and JSON.
- Trained and published a Custom Vision model and a custom NER / CLU model.
- Built an AI Search index with a skillset and run the indexer.
- Deployed an Azure OpenAI model and implemented a RAG chat endpoint.
- Configured Content Safety filters and a blocklist.
Exam-Day Time Management
With ~40-60 items in 100 minutes you have roughly 1.5-2 minutes per standalone question; case studies need more.
- Case studies: read the scenario fully, then answer every linked question — you cannot return after leaving.
- First pass (~55-65 min): lock in confident answers; flag anything over ~2 minutes.
- Second pass (~25-35 min): revisit flagged items with fresh eyes.
- Final review (~5-10 min): confirm flags and that multi-select items have every required box checked.
Highest-Yield Service-Boundary Traps
| Trap | How to Decide Correctly |
|---|---|
| Image Analysis vs Custom Vision | Pre-built tags/objects/OCR -> Image Analysis; train on your own labeled images -> Custom Vision |
| Azure AI Language vs Azure OpenAI | Structured NLP tasks (NER, sentiment, PII) -> Language; free-form generation/reasoning -> OpenAI |
| AI Search vs Document Intelligence | Index and query a corpus -> AI Search; extract fields from forms/invoices -> Document Intelligence |
| Keys vs managed identity | Production auth should use managed identity / Entra, not raw keys |
| Old names on the exam | Form Recognizer -> Document Intelligence; Cognitive Search -> AI Search; LUIS -> Conversational Language Understanding |
| Single vs multi-service resource | One billing key for many services -> multi-service; isolated billing/quota -> single-service |
Test Your Knowledge
What does the AI-102 exam primarily assess?
A
B
C
D
Test Your Knowledge
You must extract structured fields (totals, dates, line items) from scanned invoices. Which Azure service is the best fit?
A
B
C
D
Test Your Knowledge
Roughly how much time should you budget per standalone (non-case-study) question on AI-102?
A
B
C
D
Test Your Knowledge
For production authentication to an Azure AI resource, which approach does the exam favor?
A
B
C
D