Azure AI Engineer (AI-102)

Key Topics to Master:

• Service selection for generative AI, computer vision, natural language, speech, information extraction and knowledge mining
• Creating an Azure AI resource and choosing models and deployment options
• Installing and using the appropriate SDKs and APIs, and finding a service's default endpoint
• Integrating Foundry Services into a CI/CD pipeline
• Container deployment planning for AI services
• Monitoring an Azure AI resource and managing costs
• Managing and protecting account keys, and configuring authentication for a Foundry resource
• Content moderation, content filters and blocklists
• Prompt shields, harm detection and a responsible AI governance framework

Key Topics to Master:

• Deploying a hub, a project and supporting resources in Microsoft Foundry
• Implementing a prompt flow solution and using prompt templates
• Retrieval-augmented generation by grounding a model in your data
• Evaluating models and flows, and enabling tracing and feedback collection
• Provisioning an Azure OpenAI in Foundry Models resource and deploying a model
• Generating images with DALL-E and working with large multimodal models
• Parameters that control generative behaviour, including temperature and top-p
• Model monitoring, diagnostic settings, scaling and foundational model updates
• Fine-tuning a generative model and orchestrating multiple models

Key Topics to Master:

• Agent use cases and when an agent beats a single prompt or a flow
• Configuring the resources required to build an agent
• Creating an agent with the Microsoft Foundry Agent Service
• Implementing complex agents with the Microsoft Agent Framework
• Orchestration for a multi-agent solution
• Handling multiple users and autonomous capabilities in a workflow
• Testing and optimising agent behaviour
• Deploying an agent to production

Key Topics to Master:

• Selecting visual features and interpreting an image analysis response
• Object detection and image tagging
• Extracting printed and handwritten text with Azure Vision in Foundry Tools
• Choosing between image classification and object detection for a custom model
• Labelling images, training a custom model and reading its evaluation metrics
• Publishing and consuming a custom vision model, including code-first
• Azure AI Video Indexer for video and live stream insights
• Spatial Analysis for detecting the presence and movement of people in video

Key Topics to Master:

• Key phrase extraction, entity recognition, sentiment analysis and language detection
• Personally identifiable information detection in text
• Text and document translation with Azure Translator in Foundry Tools
• Text-to-speech, speech-to-text and Speech Synthesis Markup Language
• Custom speech models, intent recognition and keyword recognition
• Speech-to-speech and speech-to-text translation
• Creating intents, entities and utterances, then training, evaluating and deploying a language understanding model
• Custom question answering projects, multi-turn conversations, alternate phrasing and chit-chat
• Custom translation model training, improvement and publishing

Key Topics to Master:

• Provisioning an Azure AI Search resource, creating an index and defining a skillset
• Data sources, indexers and running an indexer
• Custom skills included in a skillset
• Query syntax including sorting, filtering and wildcards
• Knowledge Store projections as file, object and table projections
• Semantic search and vector store implementations
• Prebuilt Document Intelligence models, and training, testing and publishing custom models
• Composed document intelligence models
• OCR pipelines, document summarisation and classification, and ingesting documents, images, video and audio with Content Understanding

• 1Verify on the Microsoft exam and assessment lab retirement list that AI-102 retired on June 30, 2026 before spending money on any course
• 2Download the AI-103 study guide and map its five skill areas against the six areas of the final AI-102 blueprint
• 3Note the language shift: AI-103 expects Python development experience, where AI-102 accepted Python or C#
• 4Create an Azure subscription and a Microsoft Foundry project, and confirm your region supports the models you want
• 5Take the free practice assessment for AI-103 on Microsoft Learn and record the per-area result

• 1Create a Microsoft Foundry hub and project and deploy one model, noting quota and rate limit settings
• 2Connect to the deployed model from Python using the Foundry SDK and again with a raw REST call
• 3Switch the connection from an account key to a managed identity and confirm the keyless path works
• 4Configure a private endpoint for the AI resource and verify public access is refused
• 5Enable diagnostic settings and send resource logs to a Log Analytics workspace

• 1Configure content filters at the deployment level and test each severity threshold against sample prompts
• 2Add a blocklist and confirm which requests it stops that the default filter allows through
• 3Enable prompt shields and attempt a direct and an indirect prompt injection to see what is detected
• 4Read the Microsoft responsible AI documentation on harm categories and write down the four categories and their severity levels
• 5Run a safety evaluation against a set of prompts and read the resulting report

• 1Deploy a chat completion model and a text embedding model and compare their pricing units
• 2Write code that varies temperature, top-p, max tokens and stop sequences and observe the effect on output
• 3Generate images with DALL-E and pass an image into a multimodal model to compare the two workflows
• 4Build a prompt template with variables and reuse it across three different inputs
• 5Fine-tune a small model on a labelled dataset and compare its output with the base model

• 1Provision an Azure AI Search resource, create a data source, an index and an indexer over a document set
• 2Add a skillset with a built-in skill and then a custom skill backed by an Azure Function
• 3Configure semantic ranking and a vector field, then compare keyword, semantic and hybrid search results for the same query
• 4Ground a chat model in the index and inspect the citations it returns
• 5Configure a Knowledge Store projection and inspect the file, object and table outputs

• 1Build a single agent with the Microsoft Foundry Agent Service, giving it one tool and one knowledge source
• 2Add function calling with a schema you define, and handle the tool call round trip in code
• 3Implement conversation memory and observe how it changes multi-turn behaviour
• 4Build a two-agent orchestration where one agent delegates a subtask to another
• 5Add an approval step before a state-changing tool runs, then trace the agent's decisions

• 1Call Azure Vision in Foundry Tools with several visual feature combinations and read the JSON response field by field
• 2Extract printed and handwritten text from the same document and compare the confidence scores
• 3Train a custom image classification model and an object detection model on the same small dataset and compare precision and recall
• 4Publish a custom vision model and consume it from code rather than the portal
• 5Run a video through Azure AI Video Indexer and review the extracted insights

• 1Run key phrase extraction, named entity recognition, sentiment analysis, language detection and PII detection against one text corpus
• 2Translate a document with Azure Translator in Foundry Tools and compare it with an LLM translation of the same text
• 3Implement text-to-speech and shape the output with Speech Synthesis Markup Language, adjusting prosody and pronunciation
• 4Build a custom question answering project, add question and answer pairs, create a multi-turn prompt flow and publish it
• 5Train a conversational language understanding model with intents, entities and utterances, then evaluate and deploy it

• 1Run three prebuilt Document Intelligence models, including invoice and receipt, and read the confidence values on extracted fields
• 2Label a small document set, train a custom model, test it and publish it
• 3Build a composed model from two custom models and observe how routing works
• 4Use Azure Content Understanding in Foundry Tools to produce structured and markdown outputs from the same document
• 5Ingest an image, an audio file and a video through Content Understanding and compare the analyzer outputs

• 1Set up tracing across a generative application and read the token analytics and latency breakdown
• 2Configure quota and rate limits on a deployment and trigger a throttling response deliberately
• 3Model the monthly cost of a RAG application, separating embedding, storage, query and completion costs
• 4Add a model monitoring configuration that reports drift and grounding quality
• 5Deploy a container image of an AI service to a local or edge target

• 1Retake the Microsoft Learn practice assessment under timed conditions and compare it with your week one baseline
• 2Work through the Microsoft exam sandbox and interact with build list, hot area, drag and drop and case study question types
• 3Practise locating three specific facts on learn.microsoft.com quickly, because the exam gives access to that domain but adds no extra time
• 4Decide in advance where you would take a break, given that you cannot return to any question already viewed
• 5Rewrite every missed practice question as a one-sentence rule about which service or parameter was correct

• 1Spend the largest remaining block on generative AI and agentic solutions, which carry 30 to 35 percent of AI-103
• 2Re-read the AI-103 study guide bullet list and tick off every item you have built at least once
• 3Check the Microsoft Certification deals page for an exam and retake bundle before paying full price
• 4Confirm your Microsoft Learn profile name matches your government-issued photo identification exactly
• 5Book the exam and read Microsoft's exam duration and experience page so the seat time and break rules hold no surprises