Quel est le niveau de maturité de votre organisation Data & IA ?Faites le diagnostic
Toutes les formations

FORMATION IA

AWS Bedrock pour les Déploiements Multi-Modèles

Construisez et gouvernez des charges de travail d'IA générative en production sur AWS Bedrock avec plusieurs modèles fondamentaux.

Format
bootcamp
Durée
16–24h
Niveau
practitioner
Taille de groupe
6–16
Prix / participant
€2K–€3K
Prix groupe
€18K–€40K
Public
AWS engineers, cloud architects, and ML engineers building generative AI products on AWS infrastructure
Prérequis
Solid Python programming skills, working knowledge of AWS core services (IAM, S3, Lambda), and basic familiarity with large language model concepts

Ce qu'elle couvre

Ce programme praticien en présentiel équipe les équipes d'ingénierie orientées AWS pour sélectionner, configurer et déployer des modèles fondamentaux d'Anthropic, Meta, Cohere et Mistral via AWS Bedrock. Les participants apprennent à construire des pipelines de génération augmentée par récupération avec Bedrock Knowledge Bases, à concevoir des workflows autonomes multi-étapes avec Bedrock Agents, et à appliquer des politiques de sécurité avec Guardrails. Le programme combine des ateliers de codage en direct, des revues d'architecture et des scénarios de déploiement réels.

À l'issue, vous saurez

  • Invoke and compare multiple Bedrock foundation models programmatically using Boto3 and evaluate them against latency, cost, and quality benchmarks
  • Build a production-ready RAG application using Bedrock Knowledge Bases with an OpenSearch vector store and custom chunking strategies
  • Design and deploy a Bedrock Agent with custom action groups that integrates external APIs and executes multi-step reasoning tasks
  • Configure Bedrock Guardrails to enforce content filters, PII redaction, and topic restrictions across all deployed models
  • Apply IAM least-privilege policies, VPC endpoint configurations, and cost controls to a secure, compliant Bedrock deployment

Sujets abordés

  • Bedrock model catalogue: Anthropic Claude, Meta Llama, Cohere Command, Mistral — selection criteria and trade-offs
  • Invoking foundation models via the Bedrock API and AWS SDK (Python/Boto3)
  • Building RAG pipelines with Bedrock Knowledge Bases and Amazon OpenSearch
  • Designing multi-step agentic workflows with Bedrock Agents and custom action groups
  • Enforcing content and safety policies with Bedrock Guardrails
  • Fine-tuning and continued pre-training workflows on Bedrock
  • IAM roles, VPC endpoints, and security best practices for Bedrock deployments
  • Cost monitoring, token budgeting, and latency optimisation across models

Modalité

Delivered as a 3-day intensive bootcamp (on-site or virtual instructor-led). Approximately 60% of time is hands-on lab work within participants' own AWS accounts or a provided sandbox environment. Participants receive a pre-configured AWS CloudFormation template, a GitHub repository of reference architectures, and a post-bootcamp architecture review session. Remote delivery uses Zoom or Teams with shared VS Code Server environments. A follow-up office-hours session (2 hours) is included two weeks after delivery.

Ce qui fait que ça marche

  • Teams bring a real internal use case to the bootcamp and build against it rather than working on toy examples
  • A dedicated AWS account with Bedrock access is provisioned before day one to avoid lab delays
  • Security and compliance stakeholders join at least the Guardrails and IAM module to align on governance requirements from the start
  • Post-bootcamp, teams nominate a Bedrock internal champion responsible for maintaining the reference architecture and onboarding future colleagues

Erreurs fréquentes

  • Defaulting to a single model (usually Claude) without establishing a model selection framework, leading to cost and capability mismatches later
  • Building RAG pipelines without tuning chunking strategies or embedding models, resulting in poor retrieval relevance in production
  • Skipping Guardrails configuration until post-launch, causing compliance and reputational incidents
  • Over-provisioning model throughput units (MTUs) without understanding Bedrock's provisioned vs on-demand pricing trade-offs

Quand NE PAS suivre cette formation

This bootcamp is not appropriate for teams that have not yet chosen AWS as their cloud provider or whose organisation is still evaluating whether to build vs buy generative AI capabilities — a cloud-agnostic AI strategy workshop should come first.

Fournisseurs à considérer

Sources

Cette formation fait partie d'un catalogue Data & IA construit pour les leaders sérieux sur l'exécution. Lancez le diagnostic gratuit pour voir quelles formations sont prioritaires pour votre équipe.