Event 1

Summary Report: “GenAI & Data”

Event Objectives

  • Update on GenAI trends and strategies on AWS
  • Learn how to build a unified data foundation for AI/Analytics
  • Introduction to the AI-Driven Development Lifecycle (AI-DLC) in software development
  • Gain insights into security for GenAI applications and the role of AI Agents in enterprises

Speakers

  • Jun Kai Loke – AI/ML Specialist SA, AWS
  • Kien Nguyen – Solutions Architect, AWS
  • Tamelly Lim – Storage Specialist SA, AWS
  • Binh Tran – Senior Solutions Architect, AWS
  • Taiki Dang – Solutions Architect, AWS
  • Michael Armentano – Principal WW GTM Specialist, AWS

Key Highlights

1. Building a Unified Data Foundation on AWS for AI & Analytics

  • Strategies for constructing a unified, scalable data foundation for AI/Analytics.
  • End-to-end data pipeline: ingestion → storage → processing → access → governance.
  • Overcoming 3 silos (Data/People/Business); enable self-service with standardized governance.
  • Core services: S3, Glue, Redshift, Lake Formation, OpenSearch, Kinesis/MSK.

2. Building the Future: GenAI Adoption Strategy on AWS

  • Vision and trends of GenAI; roadmap for enterprise adoption.
  • Amazon Bedrock: model choice, customization/RAG, guardrails, cost/latency optimization.
  • AgentCore: framework-independent runtime, tool integration gateway, identity & observability.
  • Amazon Nova and ecosystem frameworks (CrewAI, LangGraph, LlamaIndex, Strands).

3. Securing Generative AI Applications with AWS

  • Risks per OWASP LLM (LLM01/02); ensuring safe output handling.
  • Security at multiple layers: infrastructure, models, applications; IAM, encryption, zero-trust, continuous monitoring.
  • 5 Security Pillars: Compliance & Governance, Legal & Privacy, Controls, Risk Management, Resilience.
  • Generative AI Security Scoping Matrix (Scope 1 → 5): from consumer apps to self-trained models.
  • Bedrock Guardrails: filter sensitive content with configurable thresholds.
  • Human-in-the-loop: human approval/intervention when needed.
  • Observability (OpenTelemetry): transparent monitoring, logging, and tracing of AI behaviors.

4. Beyond Automation: AI Agents as Productivity Multipliers

  • Agentic AI: from assistants → multi-agent systems; less human oversight, more autonomy.
  • Applications: customer support, BI with Amazon Q (QuickSight), workflow automation.
  • Amazon Q in QuickSight: Build Dashboards/Reports, Data Q&A, Executive Summaries.
  • Expected value: exponential productivity gains; requires strong data foundation & governance.

5. Reliability and Veracity of GenAI

  • Challenge of hallucination → mitigated via Prompt Engineering, RAG, Fine-tuning, Parameter Tuning.
  • RAG in action: user input → embeddings → contextual retrieval → LLM → grounded response.

6. AI-Driven Development Lifecycle (AI-DLC)

  • An AI-centric lifecycle: Inception → Construction → Operation.
  • Evolution: AI-Assisted → AI-Driven → AI-Managed; AI orchestrates, humans approve.
  • Deployment infrastructure: IaC, automated testing, observability, risk management.

7. Amazon SageMaker (Unified Studio – Next Gen)

  • One unified environment for data, analytics, and AI: SQL analytics, data processing, model development/training, GenAI app development, BI, streaming, search analytics.

  • Lakehouse + Governance: catalog/lineage, policy-based access, auditing; unified Data & AI governance.

  • Zero-ETL integration: core S3 ↔ Redshift, connections to Aurora, DynamoDB, RDS, OpenSearch, Kinesis/MSK, Salesforce, SAP, ServiceNow.

  • Full MLOps: pipelines/experiments, model registry, deployment endpoints, Feature Store, monitoring.

  • Integrated with Bedrock & JumpStart: access to foundation models (via Bedrock), reference solutions, and accelerated deployment on SageMaker.

  • SDLC automation: From planning to maintenance

  • Code transformation: Java upgrade, .NET modernization

  • AWS Transform agents: VMware, Mainframe, .NET migration

Key Takeaways

Design Mindset

  • Business-first approach: Always start from business needs, not technology.
  • Ubiquitous language: The importance of shared vocabulary between business and tech, especially in teamwork and communication with mentors.
  • Bounded contexts: Understanding how to partition domains to avoid complexity when scaling.

Architecture & Technology

  • Unified Data Foundation: ingestion → storage → processing → access → governance.
  • GenAI on AWS: Bedrock (model choice, guardrails, RAG), AgentCore (runtime, gateway, identity, observability), Nova LLMs.
  • AI Agents: from assistants → multi-agent systems; real-world use cases like customer support and BI with Amazon Q.
  • AI-DLC: AI as the core collaborator in SDLC (Inception → Construction → Operation).
  • Security-first mindset: Guardrails, human-in-the-loop, governance & monitoring (OpenTelemetry).

Strategy & Application

  • Phased approach: Avoid rushing; need a clear roadmap for modernization & AI adoption.
  • Zero-ETL & Unified Studio (SageMaker): Simplify data integration, centralize AI lifecycle management.
  • ROI measurement: Not just cost savings, but agility and productivity.

Applying to Work

  • In my project:

    • Experiment with AI Agents for workflows like registration/login or customer support.
    • Apply validation/guardrails to ensure safe integration of GenAI features.

  • In team projects (Sprint 0, serverless vs containerization):

    • Apply AI-DLC principles to split tasks logically: AI supports research/code generation, team reviews & approves.
    • Understand when to use Lambda (serverless) vs ECS/Fargate (containers).

  • In my learning path:

    • Recognize the need for a business-first approach when writing documents and gathering requirements.
    • Acknowledge that a strong data foundation is critical for any successful GenAI application.

Event Experience

  • Learned directly from AWS experts on Data, GenAI adoption, Security, AI Agents, and AI-DLC.
  • Slides and case studies gave me a clear picture of how AgentCore works in real-world scenarios.
  • Understood how AWS envisions the future of software development: AI not just as an assistant but as a core lifecycle component.
  • Realized that successful GenAI adoption requires solid data foundation + strong security + structured strategy.

Lesson learned

  • AI Agents and AgentCore will soon become critical in enterprise applications → I should learn early to stay ahead.
  • Data platform & governance are essential → not just coding, but also managing data properly.
  • AI-DLC highlights AI’s role in future SDLC → I can experiment with small projects now.
  • Security is not an afterthought; it must be built into GenAI systems from the start.

Some event photos

event-1 event-1