Event 4

Workshop Harvest Report: “AWS Cloud Mastery Series #1: GENERATIVE AI, RAG & AWS AGENTIC AI”

Event Objectives

  • Master the art of Prompt Engineering to effectively control AI models.
  • Explore the ecosystem of Pretrained AI Services available on AWS.
  • Gain a deep understanding of building AI applications using RAG (Retrieval-Augmented Generation).
  • Update on the latest trends in Agentic AI and how to move AI Agents from prototype (POC) to production using Amazon Bedrock AgentCore.
  • Explore the Pipecat Framework for building real-time voice-based virtual assistants.

Speakers

  • Lam Tuan Kiet - Sr. DevOps Engineer (FPT Software)
  • Danh Hoang Hieu Nghi - AI Engineer (Renova Cloud)
  • Dinh Le Hoang Anh - Cloud Engineer Trainee (First Cloud AI Journey)

Key Highlights

1. Prompt Engineering & Foundation Models (The Core Foundation)

Before diving into complex services, the event emphasized the importance of understanding and communicating with Foundation Models via Amazon Bedrock.

  • Zero-shot / Few-shot Prompting: Techniques involving direct instructions or providing examples to guide the model’s output format.
  • Chain of Thought (CoT): A crucial technique requiring the model to “think step-by-step,” significantly improving accuracy for complex logical problems.

2. Pretrained AWS AI Services (Ready-to-Use APIs)

Introduction to “ready-to-use” APIs that integrate intelligent features without model training:

  • Image/Video: Amazon Rekognition.
  • Language: Amazon Translate, Comprehend, Textract (OCR).
  • Audio: Amazon Polly (Text-to-Speech), Transcribe (Speech-to-Text).

3. RAG - Retrieval Augmented Generation

A process helping AI answer based on enterprise data, reducing hallucinations:

  • Embeddings: Using Amazon Titan Text Embeddings V2 to vectorise text for semantic search.
  • Knowledge Bases for Amazon Bedrock: Fully managed process handling Chunking -> Vector Store -> Retrieval -> Generation.

4. The Evolution to Agentic AI

The event introduced the next evolution of GenAI:

  1. GenAI Assistants: Follow rules, automate repetitive tasks.
  2. GenAI Agents: Goal-oriented, handling a broader range of tasks.
  3. Agentic AI Systems: Multi-agent systems acting fully autonomously with minimal human oversight.

The “Prototype to Production Chasm”: Moving Agents from POC to Production faces major hurdles regarding:

  • Performance & Scalability.
  • Security & Governance.
  • Complexity: Difficulties in managing Memory, access controls, and auditing Agent interactions.

5. Amazon Bedrock AgentCore: Bridging the Gap

To solve these challenges, AWS introduced AgentCore - a comprehensive platform for building and operating Agents:

  • Key Components:
    • Runtime & Memory: Execution environment and the ability to “remember” interaction history/learning.
    • Identity & Gateway: Identity management and secure connection gateways.
    • Code Interpreter: Allows Agents to write and execute code to process complex data.
    • Observability: Tools to monitor and audit agent activities.
  • Benefit: Allows developers to focus on business logic rather than infrastructure security or context management.

6. Pipecat: Framework for Real-time Voice AI

An interesting Open Source framework introduced for building Multimodal Virtual Assistants:

  • Focus: Optimized for Real-time interactions and conversational streaming.
  • Pipeline Mechanism:
    1. WebRTC Input: Receives audio signals from the user.
    2. STT (Speech-to-Text): Converts voice to text.
    3. LLM Processing: Processes natural language to generate a response.
    4. TTS (Text-to-Speech): Converts text back to voice.
    5. Output: Streams audio back to the user with ultra-low latency.

Event Experience & Reflection

Participating in this workshop expanded my perspective from basic concepts to the cutting-edge technologies shaping the future of AI.

1. The Shift from “Q&A” to “Action” (Agentic AI)

The most impressive concept for me was Agentic AI. Previously, I viewed AI primarily for chatting or summarization. However, through the AgentCore presentation, I see a future of “virtual employees” capable of planning, using tools (like web browsers or code interpreters), and solving complex workflows without constant human hand-holding.

2. Solving the “Production” Puzzle

I resonated deeply with the discussion on the “Chasm” between POC and Production. Tools like Amazon Bedrock AgentCore are essentially the key to building enterprise trust. They provide the necessary security layers (Identity) and control mechanisms (Observability) that allow businesses to confidently delegate tasks to AI.

3. The Potential of Voice AI with Pipecat

The Pipecat demo was fascinating. Combining WebRTC with AI models to create fluid, low-latency conversations opens up endless practical applications, such as intelligent virtual call centers, AI interview assistants, or real-time language tutors.

Conclusion

The “Generative AI & Agentic AI on AWS” workshop provided a valuable panoramic view:

  • Present: We rely on RAG and Prompt Engineering to work effectively with data.
  • Future: We are entering the era of Agentic AI, where Autonomous Agents will transform business operations.
  • Tools: With the AWS ecosystem (Bedrock, AgentCore) and Frameworks (Pipecat, LangChain), technical barriers are being removed, empowering engineers to turn breakthrough ideas into reality.

Some photos from the event