Event 6

Summary Report: “AWS Cloud Mastery Series #1: AI/ML/GenAI on AWS”

Event Objectives

  • Provide a practical and strategic overview of AI/ML and Generative AI capabilities on AWS
  • Demonstrate how foundation models and pretrained services can be integrated into enterprise solutions
  • Introduce best practices in prompt engineering, RAG (Retrieval-Augmented Generation), and orchestration for production use
  • Expose participants to AWS tools for accelerating AI adoption, including Bedrock and related services
  • Equip attendees with actionable guidance for designing scalable, responsible, and cost-effective AI systems on the cloud

Speakers

  • AWS Cloud Mastery Series instructors and engineers (AWS Vietnam office)
  • Workshop facilitators and technical demonstrators for Amazon Bedrock and SageMaker

Key Highlights

Overview of AWS AI/ML Ecosystem

The workshop began with a structured overview of AWS’s AI/ML offerings, clarifying how managed services, foundation-model hosting, and orchestration layers fit together. Attendees received a map of the stack: pretrained services for rapid prototyping, managed ML platforms for training and deployment, and foundation-model hosting for advanced GenAI scenarios.

Generative AI with Amazon Bedrock

A major portion of the session focused on Amazon Bedrock and enterprise GenAI design patterns:

  • Comparison and selection strategies for foundation models (Claude, Llama, Titan) with trade-offs in cost, latency, reasoning style, and enterprise suitability.
  • Best practices in prompt engineering: structured prompts, chain-of-thought reasoning, and few-shot techniques to improve reliability and control.
  • Retrieval-Augmented Generation (RAG) architecture: integrating knowledge bases to ground model outputs and improve factuality.
  • Bedrock Agentcore: orchestrating multi-step workflows and tool-augmented reasoning to support more complex application flows.

Pretrained AI Services and Practical Tooling

The workshop covered AWS’s pretrained services that speed development without full model training:

  • Amazon Comprehend — text analytics and NLP
  • Amazon Rekognition — image and video analysis
  • Amazon Textract — document extraction and OCR
  • Amazon Translate / Transcribe / Polly — translation, speech-to-text, and text-to-speech
  • Amazon Kendra — enterprise search
  • Amazon Personalize — personalization and recommendations

The instructors emphasized when to use pretrained services versus custom models and how to combine them with Bedrock for hybrid solutions.

Event Agenda Overview

  • 8:30 AM – 9:00 AM | Welcome & Introduction

    • Participant registration and networking
    • Workshop overview and learning objectives
    • Ice-breaker activity
    • Overview of Vietnam’s AI/ML landscape

  • 9:00 AM – 10:30 AM | AWS AI/ML Services Overview

    • Amazon SageMaker: End-to-end ML platform
    • Data preparation and labeling workflows
    • Model training, tuning, and deployment
    • Integrated MLOps capabilities
    • Live Demo: SageMaker Studio walkthrough

  • 10:30 AM – 10:45 AM | Coffee Break

  • 10:45 AM – 12:00 PM | Generative AI with Amazon Bedrock

    • Foundation Models: Claude, Llama, Titan – comparison & selection guide
    • Prompt engineering: Chain-of-thought, few-shot prompting
    • Retrieval-Augmented Generation (RAG) architecture & knowledge base design
    • Bedrock Agents for multi-step workflows and tool integrations
    • Guardrails configuration for safe and controlled AI output
    • Live Demo: Building a Generative AI chatbot using Bedrock

  • 12:00 PM | Lunch Break (Self-arranged)

Key Takeaways

Technical Patterns & Best Practices

  • Adopt a layered AI stack: pretrained services for quick wins, SageMaker for model lifecycle management, and Bedrock for foundation-model use cases.
  • RAG is essential for enterprise-grade GenAI: it grounds generation in verified documents and reduces hallucination risk.
  • Prompt engineering materially affects quality and cost — structuring prompts for concise, focused responses saves inference time and improves reliability.
  • Orchestration (Agentcore) enables multi-step reasoning and tool use, allowing models to interact with databases, APIs, and other services safely.

Operational & Governance Considerations

  • Data governance, access control, and explainability must be integrated from design time — not added later.
  • Monitor inference cost and latency trade-offs when choosing foundation models; smaller/lightweight models can be used where latency and cost matter most.
  • Implement logging, auditing, and evaluation pipelines to verify model outputs and track drift over time.

Applying to Work

  • Start small: prototype with pretrained services and a minimal RAG pipeline before investing in heavy model training.
  • Use S3 for content storage and implement a search/index (e.g., Kendra or OpenSearch) to back RAG retrieval.
  • Orchestrate multi-step flows with Agentcore or Lambda-based controllers for deterministic, auditable behavior.
  • Add CI/CD for model artifacts and prompt/version management to ensure reproducibility.
  • Define metrics for quality, latency, and cost; instrument systems with CloudWatch and regular evaluation jobs.

Event Experience

Attending the AWS Cloud Mastery Series #1 workshop offered a balanced mix of conceptual framing and practical techniques. The sessions helped translate abstract GenAI ideas into concrete engineering patterns that can be applied in real projects.

What I Learned from the Event

  • A clear mental model of the AWS AI/ML stack and when to choose each component
  • Practical prompt engineering techniques that improve output quality and reduce cost
  • How RAG pipelines and knowledge-base integration substantially improve trustworthiness of generative answers
  • Orchestration approaches for multi-step reasoning and tool integration using Bedrock Agentcore

Networking & Community

  • Connected with practitioners and AWS engineers at the Vietnam office, gaining pointers to documentation, sample repos, and follow-up labs
  • Learned about additional workshops and community resources to continue hands-on practice

The workshop provided an essential bridge between theoretical generative AI concepts and production-ready engineering patterns on AWS. It clarified practical next steps for designing scalable, responsible, and cost-aware AI systems.