1. System Architecture Overview
This document outlines the technical implementation summary for the backend of LIO Step 1: Seed & Keyword Map, including both the Preliminary Training Loop and the Keyword Precision step. The system follows a microservices architecture with event-driven communication patterns.
1.1 High-Level Architecture Diagram
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ │ │ │ │ │
│ User Interface │────▶│ API Gateway │────▶│ Orchestrator │
│ │ │ │ │ │
└─────────────────┘ └─────────────────┘ └────────┬────────┘
│
▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ │ │ │ │ │
│ Vault Service │◀───▶│ Roundtable Svc │◀───▶│ Keymaker Service│
│ │ │ │ │ │
└─────────────────┘ └─────────────────┘ └────────┬────────┘
│
▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ │ │ │ │ │
│ State Store │◀───▶│ Keyword Precision│◀───▶│ Message Queue │
│ (Vector DB) │ │ Agent Service │ │ (Kafka) │
│ │ │ │ │ │
└─────────────────┘ └─────────────────┘ └─────────────────┘
1.2 Core Components
- API Gateway: Handles external requests, authentication, and routing
- Orchestrator: Manages workflow execution and service coordination
- Keymaker Service: Central orchestrator and principal internal learner
- Roundtable Service: Ensemble of multi-model LLM APIs
- Vault Service: Adjudication panel for validation and judgment
- Keyword Precision Agent Service: Specialized processor for Step 1
- State Store: Persistent storage for system state (Vector DB)
- Message Queue: Asynchronous communication channel (Kafka/RabbitMQ)