Version : 1.0 · Date : 2026-10-21
Cours complet pour l'examen AI-103 : Developing AI Apps and Agents on Azure (certification Microsoft Certified: Azure AI Apps and Agents Developer Associate), compétences mesurées au 16 avril 2026. Un chapitre par bloc de compétences du guide d'étude, avec schémas draw.io, Mermaid, exemples Python sans clé (DefaultAzureCredential), labs, pièges d'examen et scénarios corrigés. Toutes les informations viennent de Microsoft Learn, lues le 21/10/2026. Les noms de services, fonctionnalités, rôles et menus sont en anglais, comme dans le portail et à l'examen.
Le domaine bouge vite (renommages, préversions, retraits de modèles) : relire le guide d'étude AI-103 et le Model retirement schedule avant l'examen et avant chaque lab. Une évaluation d'entraînement officielle existe sur AI Skills Navigator.
Les domaines de l'examen#
| Domaine (intitulé officiel, en anglais) | Poids | Chapitres |
|---|---|---|
| Plan and manage an Azure AI solution | 25 à 30 % | 01, 02, 03, 04 |
| Implement generative AI and agentic solutions | 30 à 35 % | 05, 06, 07 |
| Implement computer vision solutions | 10 à 15 % | 08 |
| Implement text analysis solutions | 10 à 15 % | 09 |
| Implement information extraction solutions | 10 à 15 % | 10 |
Score de réussite : 700 sur 1000. Python est le langage attendu.
Les chapitres#
Chaque chapitre suit le même plan : ce que l'examen demande (texte officiel), les concepts, les schémas, la mise en œuvre (portail, Python, CLI, IaC), les tableaux comparatifs, les labs conseillés, les pièges d'examen, des scénarios corrigés, pour aller plus loin, une fiche à retenir et les références.
Couverture des objectifs de l'examen#
Les 64 compétences du guide d'étude, chacune reliée à la section qui la traite (vérification du 21/10/2026).
Plan and manage an Azure AI solution (25 à 30 %)#
| Ch. | Compétence mesurée (guide officiel, en anglais) | Section |
|---|---|---|
| 01 | Choose an appropriate model for each task, including large language models (LLMs), small language models, multimodal models, and Foundry Tools | 2 |
| 01 | Choose the appropriate Foundry services for generative tasks, grounding, vector search, agent workflows, or multimodal processing | 3 |
| 01 | Choose an appropriate method for retrieval and indexing | 4 |
| 01 | Choose appropriate memory, tool, and knowledge integration services for agent solutions | 5 |
| 02 | Design Azure infrastructure for AI apps and agent-based solutions | 2 |
| 02 | Choose appropriate deployment options | 3 |
| 02 | Configure model and agent deployments | 4 |
| 02 | Integrate Foundry projects with continuous integration and continuous deployment (CI/CD) pipelines | 5 |
| 03 | Manage quotas, scaling, rate limits, and cost footprints for model and agent workloads | 2 |
| 03 | Monitor model performance, drift, safety events, and grounding quality | 3 |
| 03 | Monitor data ingestion quality, search index health, and relevance performance | 4 |
| 03 | Configure security, including managed identity, private networking, keyless credentials, and role policies | 5 |
| 04 | Configure safety filters, guardrails, risk detection, and content moderation | 2 et 3 |
| 04 | Apply responsible AI instrumentation, including evaluators, safety evaluations, and explanation tooling | 4 |
| 04 | Implement auditing through trace logging, provenance metadata, and approval workflows | 5 |
| 04 | Govern agent behavior with oversight modes, constraints, and tool-access controls | 6 |
Implement generative AI and agentic solutions (30 à 35 %)#
| Ch. | Compétence mesurée (guide officiel, en anglais) | Section |
|---|---|---|
| 05 | Deploy and consume LLMs, small models, code models, and multimodal models | 3 |
| 05 | Implement retrieval-augmented generation (RAG) in an application | 4 |
| 05 | Design workflows, tool-augmented flows, and multistep reasoning pipelines | 5 |
| 05 | Evaluate models and apps, including detecting fabrications, relevance, quality, and safety | 7 |
| 05 | Integrate generative workflows into applications by using Foundry SDKs and connectors | 6 |
| 05 | Configure an application to connect to a Foundry project | 2 |
| 06 | Define agent roles, goals, conversation-tracking approach, and tool schemas | 2 |
| 06 | Build agents that integrate retrieval, function-calling, and conversation memory | 3 |
| 06 | Integrate agent tools, including APIs, knowledge stores, search, content understanding, and custom functions | 4 |
| 06 | Implement orchestrated multi-agent solutions | 5 |
| 06 | Build autonomous or semiautonomous workflows with safeguards and approval flow controls | 6 |
| 06 | Integrate monitoring into deployed agents, evaluate agent behavior, and perform error analysis | 7 |
| 07 | Tune generation behavior, such as prompt engineering and adjusting model parameters | 2 |
| 07 | Implement model reflection, chain-of-thought evaluations, and self-critique loops | 3 |
| 07 | Set up observability by implementing tracing, token analytics, safety signals, and latency breakdowns | 4 |
| 07 | Orchestrate multiple models, flows, or hybrid LLM and rules engines | 5 |
Implement computer vision solutions (10 à 15 %)#
| Ch. | Compétence mesurée (guide officiel, en anglais) | Section |
|---|---|---|
| 08 | Implement a solution that generates images from text prompts and reference media | 2.2, 2.3 |
| 08 | Implement a solution that generates videos from text prompts and reference media | 3 |
| 08 | Configure image-editing workflows, including inpainting, mask‑based edits, and prompt‑driven modifications | 2.4 |
| 08 | Implement workflows to edit generated videos | 3.4 |
| 08 | Select and apply appropriate generation and editing controls provided by the platform | 2.5, 3.5 |
| 08 | Build a solution that analyzes visual context by using multimodal models | 4.1 |
| 08 | Configure apps to produce concise or detailed captions for single or multiple images | 4.2 |
| 08 | Implement a solution that enables question‑answering grounded in visual evidence | 4.3 |
| 08 | Configure generation of alt‑text and extended image descriptions aligned to accessibility guidelines | 4.4 |
| 08 | Implement visual understanding by configuring Azure Content Understanding in Foundry Tools to extract visual characteristics | 5.2 |
| 08 | Implement video analysis workflows to process and interpret video segments | 5.3 |
| 08 | Configure single‑task and pro‑mode Content Understanding pipelines | 5.4 |
| 08 | Implement solutions that identify objects, components, or regions within images or video | 6 |
| 08 | Implement filters to classify unsafe or disallowed visual content | 7.1 |
| 08 | Detect and mitigate indirect prompt injection by using embedded text in images | 7.2 |
| 08 | Enforce visual policy rules, such as applying watermarks, flagging prohibited symbols, upholding brand usage requirements, and detecting potentially inappropriate content | 7.3 |
Implement text analysis solutions (10 à 15 %)#
| Ch. | Compétence mesurée (guide officiel, en anglais) | Section |
|---|---|---|
| 09 | Implement solutions to extract entities, topics, summaries, and structured JSON outputs by using generative prompting and Foundry Tools | 2 |
| 09 | Configure detection of sentiment, tone, safety issues, and sensitive content | 3 |
| 09 | Build solutions that translate text by using Azure Translator in Foundry Tools or LLM‑powered translation flows | 4 |
| 09 | Customize language model outputs for domain tasks, such as compliance summarization and domain extraction | 5 |
| 09 | Implement workflows to convert speech to text and text to speech for agentic interactions | 6 |
| 09 | Integrate speech as an agent modality, including custom speech models | 7 |
| 09 | Enable multimodal reasoning from audio inputs | 8 |
| 09 | Translate speech into other languages by using language models and Foundry Tools | 9 |
Implement information extraction solutions (10 à 15 %)#
| Ch. | Compétence mesurée (guide officiel, en anglais) | Section |
|---|---|---|
| 10 | Ingest and index content, such as documents, images, audio, and video | 2 |
| 10 | Configure semantic search, hybrid search, and vector search for grounding | 3 |
| 10 | Implement enrichment by using custom or built-in skills for text, images, and layout | 4 |
| 10 | Configure RAG ingestion flow, including documents and using optical character recognition (OCR) | 5 |
| 10 | Connect retrieval pipelines directly to workflows and agent tools | 6 |
| 10 | Extract information by using multimodal pipelines that combine OCR, layout analysis, and field extraction | 7 |
| 10 | Produce clean, grounded representations to use with agents and RAG by using Content Understanding | 8 |
| 10 | Implement analyzers for generating structured or markdown outputs for downstream reasoning by using Content Understanding | 9 |
Organisation du dossier#
AI-103/
├── README.md ce plan de révision
├── 01-…md à 10-…md un fichier par bloc de compétences
├── images/ les schémas exportés en PNG (affichés dans les chapitres)
├── DRAWIO/ les schémas sources, modifiables dans draw.io
└── outils/ les scripts qui génèrent les schémas (moteur de GenDoc)
Régénérer les schémas d'un chapitre : python outils/schemas_NN.py [noms…] depuis ce dossier (GenDoc doit être dans ..\..\GENERATEUR-PPT).
Méthode de révision conseillée#
- Lire un chapitre en entier, schémas compris ; la sécurité réseau et identité générique est dans le cours SC-500.
- Faire ses labs conseillés dans un abonnement de test, avec des quotas et des modèles disponibles dans la région ; supprimer le groupe de ressources à la fin.
- Répondre aux scénarios du chapitre sans regarder la réponse, puis passer l'évaluation d'entraînement officielle.
- La veille de l'examen : relire les fiches à retenir et les tableaux de pièges.
- Ordre conseillé : 01 → 02 → 03 → 05 → 06 → 07 → 04 → 10 → 08 → 09 (les domaines lourds d'abord).