CertifyDeckThe official AI-103 outline has five domains spanning the lifecycle of an Azure AI solution: plan and manage (25–30%), implement generative AI and agents (30–35%), vision, text and speech (20–30% combined) and information extraction (10–15%). Knowing the weights prevents overinvesting in isolated services while neglecting agents, operations or security.
Plan and manage an Azure AI solution (25–30%)
This covers model and Foundry service selection, retrieval and indexing, memory and tools, infrastructure, deployment and CI/CD. It also tests quotas, scaling, cost, monitoring, private networking, identity and roles.
Responsible AI is part of the design: filters, guardrails, risk detection, evaluators, provenance, traces, approvals and tool-access boundaries.
Implement generative AI and agents (30–35%)
The largest domain covers model consumption, RAG, reasoning flows, evaluation, and Foundry SDKs and connectors. Agent topics include goals, memory, function calling, knowledge, tools, multi-agent orchestration and human oversight.
Operational skills include prompts and parameters, reflection evaluation, tracing, token analytics, safety signals, latency and hybrid orchestration.
Vision, text and speech (20–30% combined)
Vision includes image and video generation and editing, multimodal understanding, captions, visual question answering, accessibility, Content Understanding and indirect prompt-injection defenses. Text includes entities, summaries, structured JSON, sentiment, safety and translation.
Speech covers speech to text, text to speech, custom models, audio as an agent modality and speech translation.
Information extraction (10–15%)
You must ingest documents and media, configure semantic, hybrid and vector search, apply enrichment and build RAG ingestion with OCR. Retrieval also connects directly to flows and agent tools.
Content Understanding combines OCR, layout and field extraction and produces structured or Markdown representations for agents and RAG systems. The full AI-103 outline lists every objective with a summary.
Frequently asked questions
Which domain carries the most weight in AI-103?
Implement generative AI and agents, at 30–35%. Together with plan and manage (25–30%) it makes up more than half of the exam.
Does AI-103 include vision and speech?
Yes: vision, text and speech account for 20–30% of the exam, covering multimodal generation and understanding, text analytics, speech to text and text to speech.
How long is the AI-103 exam?
120 minutes, with a passing score of 700 out of 1,000.