Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
88 lines (69 loc) · 3.26 KB

ARCHITECTURE.md

File metadata and controls

88 lines (69 loc) · 3.26 KB

Application Architecture

Overview

EPIC.search is a document search and retrieval system using modern AI techniques including vector search and Large Language Models (LLMs).

Components Diagram

graph TB
    %% Core Application Components
    WebUI["Web UI (React App)"]
    WebAPI["Web API Python Flask Orchestrator"]
    VectorAPI["Vector API Python Flask Query Engine"]
    VectorDB["Vector Database PostgreSQL"]
    LLM["LLM Service (OLLAMA or Azure OpenAI)"]
    Embedder["Document Embedder Python Processor"]
    Client["Client Browser"]
    
    %% Core Application Flow
    Client -->|HTTPS| WebUI
    WebUI --> WebAPI
    WebAPI --> VectorAPI
    VectorAPI --> VectorDB
    WebAPI --> LLM
    Embedder -->|Populates| VectorDB
Loading

Component Descriptions

  • Web UI: React-based front-end application providing user interface
  • Web API: Flask-based orchestration layer managing request flow
  • Vector API: Specialized search engine handling vector and keyword searches
  • Vector Database: PostgreSQL database storing document vectors and metadata
  • LLM Service: AI model service (OLLAMA or Azure OpenAI) providing natural language understanding
  • Document Embedder: Processing service converting documents to vectors

Application Flow

sequenceDiagram
    participant Client as Client
    participant WebUI as Web UI
    participant WebAPI as Web API
    participant VectorAPI as Vector API
    participant VectorDB as Vector DB
    participant LLM as LLM Service
    
    Client->>WebUI: User query
    WebUI->>WebAPI: Forward query
    WebAPI->>VectorAPI: Process query
    VectorAPI->>VectorDB: Keyword search
    VectorDB-->>VectorAPI: Search results
    VectorAPI->>VectorDB: Semantic search
    VectorDB-->>VectorAPI: Search results
    VectorAPI->>VectorAPI: Rank & combine results
    VectorAPI-->>WebAPI: Return ranked results
    WebAPI->>LLM: Generate response
    LLM-->>WebAPI: RAG response
    WebAPI-->>WebUI: Return response
    WebUI-->>Client: Display to user
Loading

AI Models Configuration

Vector API Models

Purpose Model Name Description
Cross Encoder cross-encoder/ms-marco-MiniLM-L-2-v2 Used for re-ranking search results
Embeddings all-mpnet-base-v2 Used for generating vector embeddings
Keyword Processing all-mpnet-base-v2 Used for keyword extraction and processing

LLM Service Configuration

The system requires an LLM service for response generation. The deployment can use either OLLAMA or Azure OpenAI based on configuration settings:

Service Example Model Configuration Deployment Type
OLLAMA qwen2.5:0.5b Configurable via MODEL_NAME and MODEL_VERSION environment variables. Supports any Ollama-compatible model. Self-hosted on VM
Azure OpenAI gpt-3.5-turbo or gpt-4 Configurable via Azure OpenAI service endpoints and API keys. Azure managed service

Configuration Setting: The LLM_PROVIDER environment variable determines which service to use:

  • LLM_PROVIDER=ollama - Uses OLLAMA service
  • LLM_PROVIDER=azure_openai - Uses Azure OpenAI service

Note: Exactly one LLM provider must be configured and deployed for the system to function.