Study Strategies and Exam Tips for the AI-102

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.

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. Prompt flow is still an outline item, but Microsoft retires it on April 20, 2027 and points new work to Microsoft Agent Framework.

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 Content Understanding video analysis. Spatial Analysis is still listed in the outline, but Microsoft retired it on March 30, 2025.

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

Hands-On Practice Is Non-Negotiable

AI-102 tests doing, not reciting. Before exam day you should have personally:

1. Provisioned an AI services resource via portal and Azure CLI.
2. Called a Vision, Language, and OpenAI endpoint over REST with correct headers and JSON.
3. Trained and published a Custom Vision model and a custom NER / CLU model.
4. Built an AI Search index with a skillset and run the indexer.
5. Deployed an Azure OpenAI model and implemented a RAG chat endpoint.
6. Configured Content Safety filters and a blocklist.

Exam-Day Time Management

If you get 40-60 items (typical for Microsoft exams) in 100 minutes, you have roughly 1.5-2 minutes per standalone question; case studies need more.

1. Case studies: read the scenario fully, then answer every linked question — you cannot return after leaving.
2. First pass (~55-65 min): lock in confident answers; flag anything over ~2 minutes.
3. Second pass (~25-35 min): revisit flagged items with fresh eyes.
4. Final review (~5-10 min): confirm flags and that multi-select items have every required box checked.

Highest-Yield Service-Boundary Traps