Poetry RAG System

Project Overview

A modular implementation scaffold for the research project described in the design docs. The system splits responsibilities across three layers:

1. Presentation – Vue 3 + Vite SPA that captures poem input, polls task status, and renders gallery content.
2. Business Logic – Spring Boot microservice that exposes REST endpoints, normalizes task state, and dispatches compute jobs.
3. AI Microservice – FastAPI + LangChain pipeline that handles embedding lookup, prompt augmentation, and Stable Diffusion inference.

Directory layout

• frontend/ – Vite + Vue 3 code with polling wrappers, auth placeholders, and a RAG flow UI component.
• backend/ – Spring Boot project stub with REST controllers, DTOs, service layer, and database model mirroring sys_generation_task.
• ai-service/ – FastAPI-based RAG pipeline with LangChain Chain, ChromaDB wrapper, Stable Diffusion triggers, and environment-ready config.

Getting started

1. Create per-module .env files (see each module README) to map URLs, DB credentials, and GPU model weights.
2. Start the backend to expose /api/v1/poetry/visualize and /api/v1/poetry/callback endpoints.
3. Run the AI microservice (preferably under uvicorn --reload) and ensure it can reach ChromaDB/embedding storage.
4. Bootstrap the frontend via npm install/npm run dev while pointing to the backend gateway.

Next steps

• Populate the frontend with actual task polling (long-poll vs. websocket) and gallery persistence.
• Implement database schema migrations plus MyBatis-Plus mapper definitions for sys_generation_task.
• Wire up actual LangChain RetrievalQA/SequentialChain sequences, LoRA-loaded diffusion model, and blob storage uploads.