Event 1

AWS re:Invent 2025 Recap (Vietnam)

Event Details

Event Objectives

• Update on the most critical announcements from AWS re:Invent 2025 (Las Vegas).
• Deep dive into Generative AI, specifically Agentic AI and Amazon Bedrock.
• Explore new data storage and infrastructure optimizations (SageMaker, S3).
• Network with AWS Solution Architects and the local tech community.

Speakers

• Mr. Thi – Solution Architect (Topic: Generative AI & Agents)
• Mr. Tung – Speaker (Topic: OpenSearch & Agentic Search)
• Various AWS Solution Architects & Account Managers

Key Highlights

Session 1: Generative AI & Agents

• Amazon Nova Models: Introduction to the new high-performance foundation models.
• Bedrock Agents: Deep dive into Orchestration, Flows, and the new Memory and Guardrails capabilities.
• Agentic AI: Moving from simple chatbots to autonomous agents that can execute multi-step workflows.

Session 2: SageMaker Unified Studio & S3 Updates

• Unified Studio: A single IDE bringing Data Engineers, Data Scientists, and AI Engineers together, breaking down silos.
• S3 Tables: Native support for the Apache Iceberg table format within S3.
• S3 Vector: A new feature enabling native vector storage in S3, offering significant cost reduction compared to dedicated vector databases.

Session 3 & 4: Search & Multimodal AI

• OpenSearch Serverless: Integration with MCP (Model Context Protocol) and Agentic Memory.
• Nova Multimodal Embeddings: Converting video and images into vectors for search.
• Bedrock Data Automation: Automated extraction of insights from multimedia content.

Session 5: AI Infrastructure

• SageMaker HyperPod: Advanced management for large-scale GPU clusters.
• SageMaker MLflow: Full lifecycle management for Machine Learning projects.

Key Takeaways

The Future is Agentic

• Autonomous Workflows: The shift is happening from “Prompt Engineering” to “Agent Engineering.” Agents with memory can maintain context over time and perform complex tasks without constant human intervention.
• Guardrails are essential: As agents become autonomous, strict policy and security guardrails are required to ensure safe operations.

Data & Compute Optimization

• Cost Efficiency: S3 Vector is a game-changer for projects requiring vector search (like RAG) but operating on a budget. It eliminates the need for expensive, dedicated vector DBs for many use cases.
• Collaboration: SageMaker Unified Studio streamlines the workflow, allowing data preparation and model training to happen in one environment.

Applying to Work

• Project Integration (Security Platform):
  • Evaluate Bedrock Agents to automate the “vulnerability scanning” workflow (e.g., an agent that runs a scan, analyzes the log, and drafts a report).
  • Implement S3 Vector to store logs and vulnerability signatures efficiently for the project’s backend.
• Architecture Refinement: Consider adopting Cognito for user management based on discussions with SAs.
• Best Practices: Apply the “Serverless first” mindset learned from the networking sessions to keep the project infrastructure lean.

Event Experience

Attending the “AWS re:Invent 2025 Recap” at the AWS Vietnam Office was a pivotal moment in my internship.

Gaining Vision

• The sessions clarified that Agentic AI is the immediate future. Seeing the demo of a Flow Agent analyzing sales data inspired ideas for our own project’s reporting features.
• Understanding Multimodal RAG opened my eyes to possibilities beyond just text processing—essential for future phases of our Security Platform where screenshot analysis might be needed.

Technical Validation

• The introduction of S3 Vector validated our team’s need for a cost-effective storage solution. It was a direct answer to a problem we were facing regarding log analysis costs.
• Networking with Mr. Thi and other SAs helped clarify doubts about IAM Policies and Serverless best practices that I struggled with in Week 3.

Community & Networking

• The atmosphere at the AWS Office (Floor 26 & 36) was energetic, with builders and developers sharing real-world challenges.
• I had the opportunity to discuss the “TheBois” team project with industry experts and received valuable feedback on our proposed architecture.