Event 1

Summary Report: “AWS Vietnam Community Day 2026”

Event Objectives

  • Explore the latest technology trends in Generative AI (GenAI), Multi-Agent systems, and AWS cloud infrastructure optimization.
  • Gain practical insights and learn solution design experiences from leading engineers and experts in Vietnam.
  • Connect with the technology community, share soft skills, teamwork lessons, and career orientation in the AI era.

Speakers & Topics

  • Tinh Truong – Platform Engineer at GoTymeX
    • Topic: Context Is Everything: Making AI Actually Work for You
  • Pham Ng Hai Anh – G-AsiaPacific Vietnam, AWS Community Builder
    • Topic: Friendly AI Assistant w/ Amazon Q in QuickSight
  • Nguyen Tuan Thinh – DevOps Engineer, First Cloud AI Journey
    • Topic: From Edge to Origin: CloudFront as Your Foundation
  • Team VIB – LotusHacks 2026
    • Topic: 36 hrs with LotusHacks: Building UTMorpho from Idea to Reality
  • Duc Dao – Solution Architect at Cloud Kinetics
    • Topic: Non-Determinism of “Deterministic” LLM Settings
  • Vy Lam – Senior Business Systems Analyst at VPBank
    • Topic: Enterprise-Grade Multi-Agent System: The Case of Startup Credit Scoring

Key Highlights

1. Context Management in AI Applications (Speaker Tinh Truong)

  • The Importance of Context: AI functions effectively only when provided with complete context information. Context bridges the gap and minimizes hallucinations of LLMs.
  • Optimization Strategy: Prompt structuring, context window size management, and matching retrieval strategies to ensure the model yields the most accurate output.

2. AI Assistant with Amazon Q in QuickSight (Speaker Pham Ng Hai Anh)

  • Integrating GenAI into Business Intelligence: Utilizing Amazon Q in QuickSight to automate data analysis and automatically generate dashboards using natural language queries.
  • Productivity Boost: Shortens time-to-insight for businesses and facilitates quick decisions through direct natural language interaction with data.

3. Infrastructure Optimization with Amazon CloudFront (Speaker Nguyen Tuan Thinh)

  • EC2 CPU Offloading: Using CloudFront to handle data compression (reducing transfer payloads, saving up to 82% CPU load for the origin EC2 servers) and TLS handshake processing.
  • Robust Security at Edge: Leveraging AWS’s global Edge network to mitigate DDoS attacks and filter malicious traffic directly at the border before it reaches the origin servers.

4. 36-Hour Hackathon Journey at LotusHacks: Building UTMorpho (Team VIB)

  • UTMorpho Project: Building an AI Agent to generate and directly edit user interfaces on an interactive visual canvas (WYSIWYG editor) from natural language descriptions, preventing design drift from repeated re-prompts.
  • Hackathon Lessons: The most practical ideas arise from daily operational pain points. Under tight time pressure (36 consecutive hours), coordination and chemistry within the team matter more than individual skills.

5. LLM Non-Determinism in Real-world Applications (Speaker Duc Dao)

  • Non-Determinism Phenomenon: Analyzing why LLMs produce variable outputs for identical inputs even with temperature configured to 0 (caused by float calculation errors on GPUs and concurrency).
  • Control Strategy: Designing system architectures and guardrails (e.g., structured output parsing) to mitigate non-deterministic behavior in enterprise-grade applications requiring strict consistency.

6. Enterprise-Grade Multi-Agent System for Credit Scoring (Speaker Vy Lam)

  • Startup Credit Approval Challenge: Overcoming the information asymmetry between traditional banking governance standards and the dynamic, volatile nature of startup data.
  • Multi-Agent Architecture: Modeling the virtual credit committee process into specialized AI agents (Financial Agent, Compliance Agent, Risk Agent, etc.).
  • Business Value: Automating complex analysis workflows, saving up to 95% of operational time and costs while strictly maintaining compliance, security, and governance standards.

Lessons Gained

AI Application Design Mindset

  • Business-First Approach: AI applications must solve real-world business challenges (e.g., credit committee approval or automated BI analytics) rather than adopting technology for technology’s sake.
  • Multi-Agent Paradigm: Deconstructing large systems into dedicated AI agents with clear responsibilities and boundaries to handle complexity systematically.

Cloud & AI Architecture

  • Leveraging the Edge: Maximizing CDN (Amazon CloudFront) capabilities to optimize content delivery and establish a robust perimeter defense shield.
  • Enforcing Consistency: Structuring large language model output mechanisms to guarantee reliable and predictable application behaviors.

Soft Skills & Teamwork

  • Critical Communication: Smooth collaboration and transparent cross-functional alignment determine project success more than isolated individual efforts.

Key Takeaways

  • Business Aspect: Implementing a Multi-Agent model automates startup credit analysis and reduces processing time by up to 95%.
  • Technical Aspect: Understanding LLM non-determinism and optimizing Context to maximize AI prompt performance.
  • Infrastructure Aspect: Deploying CloudFront at Edge to reduce data transfer fees, offload origin EC2 CPU usage by up to 82%, and bolster DDoS protection.

Applying to Work

  • Applying context optimization techniques to improve accuracy for chatbots and AI pipelines in current workflows.
  • Implementing Amazon CloudFront caching and edge security rules to optimize web server delivery and protection.
  • Researching Multi-Agent frameworks to automate complex, multi-step business approval processes.

Event Experience

Attending AWS Vietnam Community Day 2026 was a highly valuable experience, offering a comprehensive look from theoretical design to enterprise-grade AI and Cloud deployment:

Learning from Industry Experts

  • Speakers from VPBank, GoTymeX, and Cloud Kinetics shared highly specialized, practical insights from production-grade architectures.

Exploring Advanced Solutions

  • Hands-on look at next-gen BI via Amazon Q in QuickSight, investigating LLM non-determinism under zero-temperature configurations, and studying VPBank’s Multi-Agent architecture.

Community Networking

  • Discussing directly with AWS Community Builders and industry peers, broadening professional networks, and observing standard engineering practices.

Key Takeaway

  • AI and Cloud are mutually beneficial; a great AI system requires an optimized, secure-at-edge cloud infrastructure and resilient software architectures to manage the uncertainty of LLMs.

Some event photos

Ms. Ngoc Uyen speaking Figure 1: Ms. Ngoc Uyen sharing her insights at the session

Winning the Lucky Draw Figure 2: I was fortunate enough to win the Lucky Draw prize at the event

AWS Vietnam Community Day 2026 Figure 3: Overview of AWS Vietnam Community Day 2026 event

Overall, the event not only provided technical knowledge but also helped me reshape my thinking about application design, system modernization, and cross-team collaboration.