# ai学习

**Repository Path**: flyerking/ai-learning

## Basic Information

- **Project Name**: ai学习
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-10-18
- **Last Updated**: 2026-10-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# RAG Q&A System (Single-Server Deployment)

A ready-to-use Retrieval-Augmented Generation Q&A service: upload text documents → automatic parsing, chunking and vectorization → pick any external LLM from the UI → streaming answers with source citations.

- Web UI: <http://47.92.210.193:8080>
- Default password: `rag@2025` (change `ACCESS_PASSWORD` in `.env`)
- API docs: <http://47.92.210.193:8080/api/docs>

## Stack

| Layer | Choice |
| --- | --- |
| Frontend | Vue 3 + Vite + Element Plus + Pinia (static build served by Nginx) |
| Backend | Python 3.11 + FastAPI + Uvicorn (SSE streaming) |
| Vector DB | Qdrant (single container, low memory) |
| Object storage | MinIO (S3 compatible, stores uploaded originals) |
| Metadata | MySQL 8 (`xuefei` database, tables prefixed `rag_`) |
| Embeddings | local ONNX `bge-small-zh-v1.5` (512-dim, int8, 24MB) or any OpenAI-compatible endpoint |
| Retrieval | dense (Qdrant) + BM25 (jieba) fused with RRF |
| LLM | any OpenAI-compatible API: DeepSeek / OpenAI / DashScope(Qwen) / Moonshot / Zhipu / SiliconFlow / Ollama / self-hosted |
| Proxy | Nginx (static + `/api` reverse proxy, buffering off for SSE) |

## Quick start

```bash
# on the server
cd /opt/rag-service
sudo bash deploy/install.sh
```

Then open the UI, go to **系统设置 → 大模型配置**, choose a provider (e.g. DeepSeek), paste your API Key, click **测试连接** and **保存**. Upload documents in **知识库管理**, then ask questions in **智能问答**.

Supported upload formats: `txt, md, csv, json, log, html, pdf, docx, xlsx`.

See [README.md](README.md) (Chinese) for the full guide, [docs/DEPLOY.md](docs/DEPLOY.md) for operations and troubleshooting, and [docs/API.md](docs/API.md) for the HTTP contract.