AI/ML/GenAI on AWS Workshop

Summary Report: “AI/ML/GenAI on AWS Workshop”

Event Objectives

• Understand the AI/ML landscape and AWS services ecosystem in Vietnam
• Learn end-to-end machine learning with Amazon SageMaker
• Explore Generative AI capabilities with Amazon Bedrock
• Master prompt engineering and RAG (Retrieval-Augmented Generation) techniques
• Build practical AI/ML solutions using AWS services

Event Details

• Location: AWS Vietnam Office
• Date & Time: 8:30 AM – 12:00 PM, Saturday, November 15, 2025

Speakers & Coordinators

Instructors:

• Lâm Tuấn Kiệt – Senior DevOps Engineer, FPT Software – Amazon SageMaker and ML Services Overview
• Đinh Lê Hoàng Anh – Cloud Engineer Trainee, FCAJ Swinburne University of Technology – Amazon Bedrock and AWS AI/ML Services
• Danh Hoàng Hiếu Nghị – Fresher AI Engineer, Renova Cloud – Amazon Bedrock Agent Core, live demonstrations and hands-on guidance

Coordinators:

• AWS Vietnam Community Team
• FCJ Program Leaders

Event Agenda

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

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

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

Amazon SageMaker – End-to-end ML Platform

• Data Preparation and Labeling:

  • SageMaker Data Wrangler for data preprocessing
  • Ground Truth for data labeling and annotation
  • Feature Store for feature management and reuse

• Model Training, Tuning, and Deployment:

  • Built-in algorithms and custom training scripts
  • Hyperparameter tuning with automatic model optimization
  • Model deployment options: real-time, batch, and serverless inference
  • A/B testing and multi-model endpoints

• Integrated MLOps Capabilities:

  • SageMaker Pipelines for ML workflow automation
  • Model Registry for version control and governance
  • Model Monitor for detecting data drift and model quality
  • Integration with CI/CD tools for continuous deployment

• Live Demo: SageMaker Studio walkthrough

  • Creating a notebook instance
  • Training a machine learning model
  • Deploying and testing the model endpoint

10:30 – 10:45 AM: Coffee Break

• Networking and refreshments
• Q&A with AWS experts

10:45 AM – 12:00 PM: Generative AI with Amazon Bedrock and AWS AI/ML Services

AWS AI/ML Services Overview

• Amazon Rekognition: Computer vision for image and video analysis
• Amazon Translate: Automatic text translation with neural machine translation
• Amazon Textract: Extract text and data from documents
• Amazon Transcribe: Automatic speech-to-text conversion
• Amazon Polly: Text-to-speech with natural-sounding voices
• Amazon Comprehend: Natural language processing and text analytics
• Amazon Kendra: Intelligent search service powered by ML
• Amazon Lookout: Anomaly detection for operational data
• Amazon Personalize: ML-powered recommendation engine

Foundation Models: Claude, Llama, Titan

• Model Comparison & Selection Guide:
  • Claude (Anthropic): Best for conversational AI and complex reasoning
  • Llama (Meta): Open-source flexibility and customization
  • Titan (Amazon): Cost-effective and AWS-native integration
  • Choosing the right model for your use case

Prompt Engineering Techniques

• Effective Prompting Strategies:
  • Clear instructions and context setting
  • Few-shot learning with examples
  • Chain-of-Thought (CoT) reasoning for complex tasks
  • Role-based prompting and persona definition
• Advanced Techniques:
  • Temperature and token control
  • System prompts vs user prompts
  • Prompt templates and reusability

Retrieval-Augmented Generation (RAG)

• RAG Architecture:
  • Vector databases and embeddings
  • Semantic search and document retrieval
  • Context injection into prompts
• Knowledge Base Integration:
  • Amazon Bedrock Knowledge Bases
  • Amazon S3 for storing documents and knowledge base data
  • Connecting to data sources (S3 buckets, databases, APIs)
  • Chunking strategies and metadata management
  • S3 bucket policies and access control for secure data storage

Amazon Bedrock Agent Core

• Agent Orchestration:
  • Bedrock Agent Core for building autonomous AI agents
  • Multi-step reasoning and task planning
  • Action groups and API integrations
  • Memory and conversation state management
• Tool Integrations:
  • AWS Lambda functions as tools for custom business logic
  • Lambda integration for real-time data processing
  • External API connections via Lambda
  • Database queries and data retrieval through Lambda
  • Serverless architecture benefits with Lambda + Bedrock

Guardrails: Safety and Content Filtering

• Content moderation and toxicity detection
• PII (Personally Identifiable Information) filtering
• Topic-based filtering and denied topics
• Custom guardrails for business requirements

Live Demo: Building a Generative AI Chatbot using Bedrock

• Setting up Bedrock foundation model access
• Creating a simple chatbot with prompt engineering
• Implementing RAG with Knowledge Bases
• Adding guardrails for safe responses
• Testing and iterating on the chatbot

Key Takeaways

Amazon SageMaker Capabilities

• End-to-End ML Platform: SageMaker provides all tools needed from data preparation to model deployment
• MLOps Integration: Built-in capabilities for automating and monitoring ML workflows
• Scalability: Easily scale from experimentation to production workloads
• Cost Optimization: Pay-as-you-go pricing with options for spot instances and serverless inference

Generative AI with Bedrock

• Model Diversity: Access to multiple foundation models without managing infrastructure
• Prompt Engineering: Critical skill for getting quality outputs from LLMs
• RAG Architecture: Combines the power of LLMs with your proprietary data
• Safety First: Guardrails ensure responsible AI deployment
• Agent Capabilities: Enable complex, multi-step AI workflows

Practical Implementation

• Start with Use Cases: Identify specific business problems AI/ML can solve
• Experiment and Iterate: Use SageMaker Studio for rapid prototyping
• Leverage Pre-built Models: Start with foundation models before custom training
• Implement Guardrails: Always prioritize safety and compliance
• Monitor and Optimize: Continuously track model performance and costs

Applying to Work

• Explore SageMaker: Start with SageMaker Studio free tier to experiment with ML workflows
• Build RAG Applications: Implement knowledge base integration for domain-specific chatbots
• Practice Prompt Engineering: Develop effective prompting strategies for your use cases
• Implement MLOps: Automate ML pipelines using SageMaker Pipelines
• Deploy Bedrock Agents: Create intelligent agents for automating business processes
• Ensure Compliance: Use Bedrock Guardrails to meet regulatory requirements
• Share Knowledge: Document learnings and best practices with your team

Event Experience

Attending the “AI/ML/GenAI on AWS Workshop” at AWS Vietnam Office was an immersive learning experience that provided hands-on exposure to cutting-edge AI/ML technologies. Key experiences included:

Learning from AWS Experts

• AWS Solutions Architects provided comprehensive insights into SageMaker’s end-to-end ML capabilities
• AWS GenAI Specialists demonstrated practical applications of Amazon Bedrock and foundation models
• Real-world use cases illustrated how Vietnamese companies are leveraging AWS AI/ML services
• Expert guidance on choosing the right tools and models for specific business needs

Hands-on Demonstrations

• Witnessed SageMaker Studio in action, from data preparation to model deployment
• Saw how Amazon Bedrock enables rapid development of GenAI applications without infrastructure management
• Learned practical prompt engineering techniques that immediately improve LLM outputs
• Explored RAG architecture for building knowledge-aware AI applications
• Understood how Bedrock Agents orchestrate complex multi-step workflows

Understanding AI/ML Landscape

• Gained insights into the AI/ML adoption trends in Vietnam
• Learned about the differences between traditional ML and Generative AI approaches
• Understood when to use SageMaker vs Bedrock for different use cases
• Discovered the importance of MLOps for production ML systems

Networking and Community Building

• Connected with fellow developers and data scientists exploring AWS AI/ML services
• Exchanged ideas about practical AI/ML implementation challenges and solutions
• Built relationships with AWS experts for ongoing support and guidance
• Joined the AWS AI/ML community for continuous learning

Practical Insights Gained

• Foundation models democratize access to powerful AI capabilities without requiring massive resources
• Prompt engineering is a critical skill that significantly impacts GenAI application quality
• RAG architecture solves the problem of LLMs lacking domain-specific knowledge
• Guardrails are essential for responsible and compliant AI deployment
• SageMaker provides a complete platform that accelerates ML development and deployment

Next Steps

• Begin experimenting with SageMaker Studio using the free tier
• Build a proof-of-concept RAG application using Bedrock Knowledge Bases
• Practice prompt engineering techniques on different foundation models
• Explore Bedrock Agents for automating business workflows
• Implement MLOps practices using SageMaker Pipelines
• Continue engaging with the AWS AI/ML community for ongoing learning

Event Pictures

Overall, this workshop provided a comprehensive introduction to AWS AI/ML services, from traditional machine learning with SageMaker to cutting-edge Generative AI with Bedrock. The hands-on demonstrations and expert guidance made complex concepts accessible and immediately applicable. The key takeaway is that AWS provides a complete ecosystem for building, deploying, and scaling AI/ML applications, making it easier than ever to bring AI innovations to production.