top of page

AI-103 practice exam

Developing AI Apps and Agents on Azure

Timed practice tests for AI-103 on the real exam's time limit and blueprint. Every question is original and every option is explained. Your score is broken down by the sections Microsoft publishes, so you see what to study next. The question bank holds hundreds of practice questions.

Exam facts

  • Time limit: 120 minutes [1]
  • Passing score: 700 or greater [2]
  • Exam languages (real exam): English, Chinese (Simplified), Chinese (Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil), Spanish [1]

Tokens: 1 for $4.99, 5 for $14.99, 10 for $24.99 or 25 for $49.99. One token runs a full timed AI-103 practice exam, and every shorter session on that exam is then free. The Unlimited pass is $149.99 for a year of every practice session on every exam.

Get tokens

Exam sections and weights

Microsoft publishes these sections and objectives for AI-103, quoted here as Microsoft words them. [2]

Section 1: Plan and manage an Azure AI solution, 25 to 30% of the exam

  • "Choose an appropriate model for each task, including large language models (LLMs), small language models, multimodal models, and Foundry Tools"
  • "Choose the appropriate Foundry services for generative tasks, grounding, vector search, agent workflows, or multimodal processing"
  • "Choose an appropriate method for retrieval and indexing"
  • "Choose appropriate memory, tool, and knowledge integration services for agent solutions"
  • "Design Azure infrastructure for AI apps and agent-based solutions"
  • "Choose appropriate deployment options"
  • "Configure model and agent deployments"
  • "Integrate Foundry projects with continuous integration and continuous deployment (CI/CD) pipelines"
  • "Manage quotas, scaling, rate limits, and cost footprints for model and agent workloads"
  • "Monitor model performance, drift, safety events, and grounding quality"
  • "Monitor data ingestion quality, search index health, and relevance performance"
  • "Configure security, including managed identity, private networking, keyless credentials, and role policies"
  • "Configure safety filters, guardrails, risk detection, and content moderation"
  • "Apply responsible AI instrumentation, including evaluators, safety evaluations, and explanation tooling"
  • "Implement auditing through trace logging, provenance metadata, and approval workflows"
  • "Govern agent behavior with oversight modes, constraints, and tool-access controls"

Section 2: Implement generative AI and agentic solutions, 30 to 35% of the exam

  • "Deploy and consume LLMs, small models, code models, and multimodal models"
  • "Implement retrieval-augmented generation (RAG) in an application"
  • "Design workflows, tool-augmented flows, and multistep reasoning pipelines"
  • "Evaluate models and apps, including detecting fabrications, relevance, quality, and safety"
  • "Integrate generative workflows into applications by using Foundry SDKs and connectors"
  • "Configure an application to connect to a Foundry project"
  • "Define agent roles, goals, conversation-tracking approach, and tool schemas"
  • "Build agents that integrate retrieval, function-calling, and conversation memory"
  • "Integrate agent tools, including APIs, knowledge stores, search, content understanding, and custom functions"
  • "Implement orchestrated multi-agent solutions"
  • "Build autonomous or semiautonomous workflows with safeguards and approval flow controls"
  • "Integrate monitoring into deployed agents, evaluate agent behavior, and perform error analysis"
  • "Tune generation behavior, such as prompt engineering and adjusting model parameters"
  • "Implement model reflection, chain-of-thought evaluations, and self-critique loops"
  • "Set up observability by implementing tracing, token analytics, safety signals, and latency breakdowns"
  • "Orchestrate multiple models, flows, or hybrid LLM and rules engines"

Section 3: Implement computer vision solutions, 10 to 15% of the exam

  • "Implement a solution that generates images from text prompts and reference media"
  • "Implement a solution that generates videos from text prompts and reference media"
  • "Configure image-editing workflows, including inpainting, mask-based edits, and prompt-driven modifications"
  • "Implement workflows to edit generated videos"
  • "Select and apply appropriate generation and editing controls provided by the platform"
  • "Build a solution that analyzes visual context by using multimodal models"
  • "Configure apps to produce concise or detailed captions for single or multiple images"
  • "Implement a solution that enables question-answering grounded in visual evidence"
  • "Configure generation of alt-text and extended image descriptions aligned to accessibility guidelines"
  • "Implement visual understanding by configuring Azure Content Understanding in Foundry Tools to extract visual characteristics"
  • "Implement video analysis workflows to process and interpret video segments"
  • "Configure single-task and pro-mode Content Understanding pipelines"
  • "Implement solutions that identify objects, components, or regions within images or video"
  • "Implement filters to classify unsafe or disallowed visual content"
  • "Detect and mitigate indirect prompt injection by using embedded text in images"
  • "Enforce visual policy rules, such as applying watermarks, flagging prohibited symbols, upholding brand usage requirements, and detecting potentially inappropriate content"

Section 4: Implement text analysis solutions, 10 to 15% of the exam

  • "Implement solutions to extract entities, topics, summaries, and structured JSON outputs by using generative prompting and Foundry Tools"
  • "Configure detection of sentiment, tone, safety issues, and sensitive content"
  • "Build solutions that translate text by using Azure Translator in Foundry Tools or LLM-powered translation flows"
  • "Customize language model outputs for domain tasks, such as compliance summarization and domain extraction"
  • "Implement workflows to convert speech to text and text to speech for agentic interactions"
  • "Integrate speech as an agent modality, including custom speech models"
  • "Enable multimodal reasoning from audio inputs"
  • "Translate speech into other languages by using language models and Foundry Tools"

Section 5: Implement information extraction solutions, 10 to 15% of the exam

  • "Ingest and index content, such as documents, images, audio, and video"
  • "Configure semantic search, hybrid search, and vector search for grounding"
  • "Implement enrichment by using custom or built-in skills for text, images, and layout"
  • "Configure RAG ingestion flow, including documents and using optical character recognition (OCR)"
  • "Connect retrieval pipelines directly to workflows and agent tools"
  • "Extract information by using multimodal pipelines that combine OCR, layout analysis, and field extraction"
  • "Produce clean, grounded representations to use with agents and RAG by using Content Understanding"
  • "Implement analyzers for generating structured or markdown outputs for downstream reasoning by using Content Understanding"

Sample questions

Three questions from the AI-103 practice bank, with the answer and why each option is right or wrong.

Question 1

You are building a semantic search feature in Microsoft Foundry. You need embeddings with fewer than 1,536 dimensions per vector to cut index storage. Retrieval quality must stay at least as good as text-embedding-ada-002. Which model and setting should you use?

  • A. text-embedding-3-small with its default output dimensions
  • B. text-embedding-ada-002 with its default output dimensions
  • C. text-embedding-ada-002 with the `dimensions` parameter set to 512
  • D. text-embedding-3-small with the `dimensions` parameter set to 512

Answer: D. Only the third generation embedding models let you shrink vectors with the `dimensions` parameter. OpenAI's MTEB testing found they keep slightly better results than ada-002 even below 1,536 dimensions. [3]

  • A. Wrong. text-embedding-3-small returns 1,536 dimensions by default. Without the `dimensions` parameter, storage per vector doesn't shrink.
  • B. Wrong. text-embedding-ada-002 returns 1,536 dimensions, so the vectors don't drop below the size the requirement sets.
  • C. Wrong. The `dimensions` parameter is a feature of the third generation embedding models, so text-embedding-ada-002 can't shrink its vectors below 1,536 dimensions.
  • D. Right. Third generation embedding models accept a `dimensions` parameter. OpenAI's MTEB testing found they stay slightly better than ada-002 even below 1,536 dimensions.

Question 2

You have an app that calls a reasoning model deployment through the Chat Completions API. You need to log how many hidden reasoning tokens each request consumed, apart from the visible answer. Which response field should the app read?

  • A. `usage.completion_tokens_details.reasoning_tokens`
  • B. `usage.output_tokens_details.reasoning_tokens`
  • C. `usage.prompt_tokens_details.cached_tokens`
  • D. `usage.total_tokens`

Answer: A. Reasoning tokens never appear in the message content, yet they use context space and bill as output. Chat Completions reports them in completion_tokens_details.reasoning_tokens; the Responses API reports them in output_tokens_details.reasoning_tokens. [4]

  • A. Right. In a Chat Completions response, the reasoning token count sits under completion_tokens_details, apart from the visible output.
  • B. Wrong. output_tokens_details is where a Responses API response reports reasoning tokens. This app calls the Chat Completions API, which reports them elsewhere.
  • C. Wrong. This field reports input tokens served from the prompt cache. It holds no count of the tokens the model spent reasoning.
  • D. Wrong. The total sums input and output tokens. You can't tell the reasoning tokens apart from the visible answer with that one number.

Question 3

You are building a poster generator in a Microsoft Foundry project. You need each poster generated at 3840 x 2160 pixels from a text prompt. The solution must not rely on any output resolution that's marked experimental. Which model meets the requirement?

  • A. GPT-Image-2.5-Flare capped at 2560x1440
  • B. GPT-Image-2.5-Flare
  • C. GPT-Image-2.5-Sunburst
  • D. GPT-Image-2

Answer: D. Check the size rules per model. GPT-Image-2 is listed as best for high-resolution and 4K generation, while the GPT-Image-2.5 models treat sizes above 2560x1440 as experimental. [5]

  • A. Wrong. The cap keeps Flare off its experimental sizes, but the posters must come out at 3840 x 2160, so a 2560x1440 image misses the size requirement.
  • B. Wrong. GPT-Image-2.5-Flare accepts custom sizes, but its resolutions above 2560x1440 are experimental.
  • C. Wrong. GPT-Image-2.5-Sunburst accepts custom sizes, but its resolutions above 2560x1440 are experimental too.
  • D. Right. GPT-Image-2 takes arbitrary sizes with a long edge up to 3,840 px and up to 8,294,400 pixels, and 3840 x 2160 is a multiple of 16 on both edges.

Practice for AI-103 on ExamSens.ai

ExamSens.ai is not authorized, endorsed or sponsored by Microsoft.

References

[1] https://learn.microsoft.com/en-us/credentials/certifications/exams/ai-103/

[2] https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ai-103

[3] https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure

[4] https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/reasoning

[5] https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/dall-e

bottom of page