K-Food RAG: AI-Powered Culinary Discovery

Overview

K-Food RAG is a production-ready system designed to revolutionize how users discover and explore Korean cuisine. By integrating Retrieval-Augmented Generation (RAG) with a robust microservices architecture, it provides intelligent, context-aware search results that go beyond simple keyword matching.

Key Features

🧠 Intelligent Search

• Semantic Understanding: Uses vector embeddings to understand the intent behind user queries, not just the keywords.
• Context-Aware Results: Retrieves relevant culinary information based on ingredients, flavor profiles, and cultural context.

🛡️ Resilient Architecture

• Circuit Breakers: Protects external API integrations (like embedding services) from cascading failures.
• Queue-Based Processing: Ensures reliable data synchronization and heavy processing tasks using Bull and Redis.

📊 Comprehensive Monitoring

• Data Quality Service: Continuously monitors the consistency between the primary database and the vector store.
• Real-time Metrics: Tracks system health, sync status, and performance metrics.

RAG Implementation

The core of the system is its advanced RAG implementation, which ensures high availability and data consistency.

RAG Workflow

The system follows a streamlined Retrieval-Augmented Generation process to deliver accurate and context-aware responses.

flowchart TD
    User[User Query]:::entity --> API[API Gateway]:::service
    API --> Embed[Embedding Service]:::service
    Embed -->|Generate Vector| GenAI[Google GenAI]:::external
    GenAI -->|Vector| Embed
    
    Embed --> Search[Vector Search]:::service
    Search -->|Query Vector| Qdrant[Qdrant Vector Store]:::external
    Qdrant -->|Relevant IDs| Search
    
    Search --> Context[Context Retrieval]:::service
    Context -->|Fetch Details| DB[PostgreSQL]:::external
    DB -->|Food Data| Context
    
    Context --> LLM[LLM Processing]:::service
    LLM -->|Prompt + Context| GenAI
    GenAI -->|Natural Response| LLM
    LLM -->|Final Answer| User
    
    classDef service fill:#007bff,stroke:#0056b3,stroke-width:2px,color:#fff
    classDef entity fill:#28a745,stroke:#1e7e34,stroke-width:2px,color:#fff
    classDef external fill:#fd7e14,stroke:#e8590c,stroke-width:2px,color:#fff

Queue-Based Processing Architecture

To handle high loads and ensure data consistency, K-Food uses a dual-queue system for processing RAG operations.

graph LR
    subgraph "Queue System"
        A[rag-sync]:::external --> B[Primary Processing]:::service
        C[rag-sync-retry]:::external --> D[Retry Processing]:::service
    end
    
    subgraph "Processors"
        E[RagSyncProcessor]:::service --> B
        F[RagRetryProcessor]:::service --> D
    end
    
    subgraph "Workers"
        G[Worker 1]:::service --> E
        H[Worker 2]:::service --> E
        I[Worker 3]:::service --> F
    end
    
    classDef service fill:#007bff,stroke:#0056b3,stroke-width:2px,color:#fff
    classDef external fill:#fd7e14,stroke:#e8590c,stroke-width:2px,color:#fff

System Architecture

The application is built on a modern stack ensuring scalability and performance.

• Frontend: Next.js 16 with React 19, utilizing React Query for state management and TailwindCSS for styling.
• Backend: NestJS 11 providing a robust server-side architecture.
• Database: PostgreSQL for relational data and Qdrant for vector embeddings.
• AI Integration: Google GenAI (Gemini) for generating embeddings and natural language responses.
• Infrastructure: Redis for caching and job queues, with cloud storage integration (AWS S3/GCP).