Vertex AI Guide

Vertex AI 핵심 아키텍처 구성 요소 Vertex AI Core Architecture Components

GCP 콘솔 접속:GCP Console Access: Vertex AI Overview

Vertex AI는 데이터 준비부터 머신러닝 모델의 학습, 배포, 모니터링까지 전체 엔드투엔드 파이프라인을 지원하는 완전 관리형 AI 플랫폼입니다. MLOps 및 LLMOps 워크플로우를 구성하는 세 가지 주요 인프라 컴포넌트가 있습니다. Vertex AI is a fully managed AI platform that supports the entire end-to-end pipeline, from data preparation to machine learning model training, deployment, and monitoring. It consists of three main infrastructure components that form MLOps and LLMOps workflows.

Vertex AI Agent Applications

GCP 콘솔 접속:GCP Console Access: Agent Builder & Applications

사용자 의도를 파악하고, 엔터프라이즈 데이터베이스나 외부 API와 연동해 다중 턴(Multi-turn) 과업을 수행하는 자율형 에이전트(Autonomous Agent) 개발 아키텍처입니다. An architecture for developing autonomous agents that understand user intent and integrate with enterprise databases or external APIs to execute multi-turn tasks.

graph TD
    A[Vertex AI Platform] --> B[Search Engine];
    A --> C[Agent Engine];

    subgraph "Information Retrieval"
        B --> D[Unstructured Search];
        B --> E[Recommendation Engine];
        D -.-> D1(Website/Cloud Storage/PDF/JSON);
        E -.-> E1(Retail/Media Personalization);
    end

    subgraph "Task Execution"
        C --> F[Chatbots];
        C --> G[Autonomous Agents];
        F -.-> F1(DialogFlow CX Integration);
        G -.-> G1(API/Database Action Execution);
    end

    style A fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff
    style B fill:#334155,stroke:#475569,color:#fff
    style C fill:#334155,stroke:#475569,color:#fff
    style D fill:#475569,stroke:none,color:#e2e8f0
    style E fill:#475569,stroke:none,color:#e2e8f0
    style F fill:#475569,stroke:none,color:#e2e8f0
    style G fill:#475569,stroke:none,color:#e2e8f0
    style D1 fill:none,stroke:#64748b,stroke-dasharray: 5 5,color:#94a3b8
    style E1 fill:none,stroke:#64748b,stroke-dasharray: 5 5,color:#94a3b8
    style F1 fill:none,stroke:#64748b,stroke-dasharray: 5 5,color:#94a3b8
    style G1 fill:none,stroke:#64748b,stroke-dasharray: 5 5,color:#94a3b8
            

Vertex AI Search

GCP 콘솔 접속:GCP Console Access: Vertex AI Search

기존 키워드 기반 시스템과 의미론적 검색(Semantic Search)을 결합하여, 기업 내 흩어진 데이터를 자동으로 색인하고 지능적인 질의응답을 제공하는 완전 관리형 엔터프라이즈 검색 엔진 아키텍처입니다. A fully managed enterprise search engine architecture that combines traditional keyword systems with semantic search to automatically index scattered enterprise data and provide intelligent Q&A.

graph TD
    A[End User] -->|Query| B{Vertex AI Search API};
    
    subgraph "Indexing & Processing Pipeline"
        C1(Websites) -->|Crawling| D[Data Ingestion];
        C2(GCS / Cloud Storage) -->|Unstructured Data| D;
        C3(BigQuery / Cloud SQL) -->|Structured Data| D;
        
        D --> E[Document Parsing & Chunking];
        E --> F[Embedding Model];
        F --> G[(Vector Index)];
        E --> H[(Keyword/Lexical Index)];
    end
    
    B -->|Search| G;
    B -->|Search| H;
    
    subgraph "Retrieval & Ranking"
        G -.-> I[Hybrid Search Engine];
        H -.-> I;
        I --> J[LLM Reranker];
    end
    
    J -->|Top-K Results| K[Gemini Model];
    K -->|Generates Answer with Citations| B;

    style A fill:#1e293b,stroke:#e2e8f0,color:#fff
    style B fill:#2d3748,stroke:#60a5fa,stroke-width:2px,color:#fff
    style D fill:#475569,stroke:none,color:#e2e8f0
    style E fill:#475569,stroke:none,color:#e2e8f0
    style F fill:#475569,stroke:none,color:#e2e8f0
    style G fill:#334155,stroke:#475569,color:#fff
    style H fill:#334155,stroke:#475569,color:#fff
    style I fill:#374151,stroke:#60a5fa,stroke-dasharray: 5 5,color:#fff
    style J fill:#374151,stroke:#60a5fa,stroke-dasharray: 5 5,color:#fff
    style K fill:#2d3748,stroke:#2dd4bf,stroke-width:2px,color:#fff
            

Vertex AI RAG Engine

GCP 콘솔 접속:GCP Console Access: Vertex AI RAG Engine

환각(Hallucination) 현상을 원천적으로 차단하기 위해, 자체 언어 모델 외부에 기업의 독점적인(Proprietary) 정보 저장소(Corpus)를 구축하여 LLM 추론 과정에 실시간으로 매핑하는 아키텍처입니다. Search와 유사하나 개발자 친화적인 세밀한 파이프라인 제어를 허용합니다. To fundamentally prevent hallucinations, this architecture builds a proprietary enterprise information repository (Corpus) external to the language model, mapping it into the LLM inference process in real-time. Similar to Search but allows for fine-grained, developer-friendly pipeline control.

graph LR
    A(User Query) --> B[Embedding API];
    B --> C[Vertex Vector Search];
    
    subgraph "Knowledge Corpus (Data Sync)"
        G(Private Docs/Data) --> H[Chunking Strategy];
        H --> I[Embedding API];
        I -->|Upsert Embeddings| C;
    end

    C -.->|Retrieve KNN Docs| D[Context Assembler];
    A --> D;
    
    D -->|Prompt + Context| E[Gemini Models];
    E --> F((Grounded Response
w/ Citations));
    
    style A fill:#1e293b,stroke:#e2e8f0,color:#fff
    style B fill:#334155,stroke:#475569,color:#fff
    style C fill:#334155,stroke:#475569,color:#fff
    style D fill:#374151,stroke:#60a5fa,stroke-dasharray: 5 5,color:#fff
    style E fill:#2d3748,stroke:#60a5fa,stroke-width:2px,color:#fff
    style F fill:#1e293b,stroke:#2dd4bf,stroke-width:2px,color:#fff
    style G fill:#475569,stroke:none,color:#e2e8f0
    style H fill:#475569,stroke:none,color:#e2e8f0
    style I fill:#475569,stroke:none,color:#e2e8f0
            

BigQuery 기반 AI 연동 및 데이터 에이전트 BigQuery AI Integration & Data Agents

GCP 콘솔 접속:GCP Console Access: BigQuery SQL Workspace

데이터 이동 없는(Zero-ETL) In-database ML 추론부터 최신 Conversational Analytics(대화형 분석 에이전트)까지, 데이터 웨어하우스 자체를 지능형 분석 플랫폼으로 확장하는 최신 아키텍처 패턴입니다. A modern architecture pattern expanding the data warehouse itself into an intelligent analytics platform—ranging from Zero-ETL In-database ML inference to the latest Conversational Analytics agents.

graph TD
    subgraph "BigQuery Environment"
        A[Petabytes Structured Data] --> B{SQL Query Execution};
        A -.-> J[BigQuery Agents 
Preview];
        B -->|AI.GENERATE_TEXT / TABLE| C[LLM Remote Connection];
        B -->|AI.EMBED & SIMILARITY| D[Embedding Remote Connection];
        J -->|Conversational NL2SQL| B;
        C --> E[In-place Analysis & Schema Extraction];
        D --> F[In-place Vectorization & Search];
        E --> G[(Target BQ Table)];
        F --> G;
    end
    
    subgraph "Vertex AI Platform"
        C -.-> H(Gemini Models);
        D -.-> I(Text/Multimodal Embeddings);
    end
    
    style A fill:#1e293b,stroke:#e2e8f0,color:#fff
    style B fill:#2d3748,stroke:#60a5fa,stroke-width:2px,color:#fff
    style C fill:#334155,stroke:#475569,color:#fff
    style D fill:#334155,stroke:#475569,color:#fff
    style E fill:#475569,stroke:none,color:#e2e8f0
    style F fill:#475569,stroke:none,color:#e2e8f0
    style G fill:#1e293b,stroke:#2dd4bf,stroke-width:2px,color:#fff
    style H fill:#1e293b,stroke:#a855f7,stroke-dasharray: 5 5,color:#fff
    style I fill:#1e293b,stroke:#a855f7,stroke-dasharray: 5 5,color:#fff
    style J fill:#4c1d95,stroke:#c084fc,stroke-width:2px,color:#fff
            

Industry Use Cases (POC)

GCP 콘솔 접속:GCP Console Access: Vertex AI Workspace

다양한 산업군에서 발생한 엔터프라이즈 데이터 병목 및 검색 한계를 Vertex AI 생태계를 활용하여 해결한 개념 증명(POC) 아키텍처 사례입니다. Proof of Concept (POC) architecture use cases resolving enterprise data bottlenecks and search limitations across various industries using the Vertex AI ecosystem.

AI 기반 검색 및 분석

지능형 자동화 및 챗봇

멀티모달 응용

데이터 플랫폼 현대화