Azure AI-300: Operationalizing ML and GenAI Solutions

Understand how Azure ML workspace resources, associated resources, identities, storage, registries, compute, endpoints, and Foundry projects relate.

Understand why production AI systems need versioned code, data references, environments, prompts, model artifacts, and deployment definitions.

Create and organize Azure ML workspaces, configure associated resources, and reason about workspace boundaries.

Understand how data assets, datastores, compute targets, environments, components, and registries make ML work reproducible.

Understand how managed identity, RBAC, network isolation, Bicep, Azure CLI, and GitHub Actions support repeatable infrastructure.

Turn exploratory notebook work into tracked, reproducible experiments with parameters, metrics, artifacts, and run lineage.

Use MLflow tracking and model registry operations to turn experiment outputs into governable model artifacts.

Choose between custom training scripts, AutoML, hyperparameter tuning, distributed training, and feature retrieval.

Make preprocessing, training, evaluation, registration, and promotion into a repeatable pipeline.

Choose between managed online endpoints and batch endpoints based on scaling, latency, cost, and invocation patterns.

Operate production inference with deployment variants, traffic routing, rollback, logs, and failure triage.

Monitor model quality, service health, latency, errors, and operational telemetry after deployment.

Detect data drift, concept drift, and quality degradation, and define retraining or rollback triggers.

Integrate responsible AI evaluation into model promotion, risk review, and production readiness.

Structure GenAI applications with Foundry project environments, model deployments, identities, RBAC, and environment separation.

Choose, deploy, version, and operate foundation models including throughput and cost constraints.

Turn prompts into deployable artifacts with versioning, variants, review, automated evaluation, and rollout controls.

Design evaluation datasets, map input/output columns, and select metrics for answer quality.

Use risk/safety evaluators and custom evaluators to test product-specific failure modes.

Use traces, latency, throughput, failures, token usage, and cost telemetry for GenAI debugging and operations.

Tune chunk size, overlap, metadata, similarity thresholds, and retrieval strategy for grounded answer quality.

Choose between vector, keyword, hybrid search, semantic ranking, and embedding model changes.

Decide when to optimize prompts, retrieval, embeddings, fine-tuning data, or fine-tuned model operations.