AI-103 on Microsoft Learn: study guide, practice assessment and labs

The AI-103 study guide includes the audience profile, the five domains with their weights and the bullets that illustrate how each skill is assessed. Read it in full once at the start and use it as a checklist at the end: if a bullet sounds new, that is your next lab. In it Microsoft notes that the English version of the exam is updated first and localized versions about eight weeks later.

The exam sandbox shows the real interface and question types with generic content; give it ten minutes so you do not lose time on exam day. Official training, both self-paced learning paths and the instructor-led course, is linked from the preparation section of the exam page on Microsoft Learn. The paths are free and include guided exercises; the AI-103 outline breakdown helps you split the hours.

The certification page says the practice assessment is available on AI Skills Navigator, with an upgraded experience, and that you must be signed in to launch it. It is free, written by Microsoft, and works best as a diagnosis: take it once at the start, note your weak domains, and again halfway through.

Its limits: the question set is finite and repeats, so after two or three passes you are measuring memory, not level. For daily work you need more volume and an explanation per question. The AI-103 practice test gives you tests by domain, explanations with documentation references and timed 120-minute mock exams; free to start, no sign-up.

Plan and manage: create a Foundry project, deploy a model, connect from Python with managed identity instead of keys, set a rate limit and look at the usage metrics. Generative AI and agents: build RAG with a hybrid index in Azure AI Search and citations in the answer; then an agent with a tool (a function or an API) and conversation memory; run a relevance and fabrication evaluation and review the traces.

Vision: generate and edit an image with a mask, describe an image with a multimodal model and try a Content Understanding analyzer. Text and speech: extract entities and structured JSON with a prompt, translate a text and run speech to text and text to speech. Extraction: ingest a scanned PDF with OCR and layout analysis, index the result and use it as the agent's source. On a free or pay-as-you-go subscription each lab costs cents if you delete the resources when you finish.

Week 1: study guide, sandbox and the practice assessment as a diagnosis. Weeks 2 to 6: official training block by block, with the lab for each domain and practice questions from that domain when you finish it. Last two weeks: timed mock exams, reviewing mistakes against the documentation, and a second pass of the practice assessment. The eight-week AI-103 plan lays it out.

About the GitHub lab repositories that show up in search: the exercises in the official Microsoft Learn paths link to their own repositories, which are the reliable ones and are kept current with Foundry changes. Be wary of old collections built for AI-102: many services and SDKs have changed names and usage.