ReflectionRAG
基于 LangGraph 构建的自适应 RAG 智能问答系统,集成 通义千问 (Qwen) 大模型、Qdrant 向量数据库与 Tavily 网络搜索,实现文档解析 → 智能分块 → 向量化存储 → 问题路由 → 检索增强生成的全链路闭环。
项目特点
- 🔀 智能路由:LLM 自动判断走本地向量库检索还是网络搜索
- 🔄 自适应反馈闭环:检索文档不相关时自动重写 Query 并重新检索
- ✅ 文档相关性评分:LLM 对每篇检索文档进行 yes/no 评分,过滤噪声
- 🛡️ 检索失败降级:检索或搜索失败时自动切换原始大模型直接回答
- 📄 多格式文档解析:基于 MarkItDown 统一处理 PDF/Word/Excel/PPT
- 🧠 语义感知分块:Markdown 标题层级识别 + Token 估算自适应分块
- 📊 丰富元数据:每个 Chunk 附带章节路径、来源文件和位置信息
工作流
用户问题
│
▼
┌──────────┐
│ 问题路由 │ ◄── LLM 判断:本地 RAG 还是网络搜索
└────┬─────┘
│
┌──┴──────────┐
▼ ▼
┌──────┐ ┌──────────┐
│本地检索│ │ 网络搜索 │ ◄── Tavily Search API
└──┬───┘ └────┬─────┘
│ │
▼ │
┌──────────┐ │
│ 文档评分 │ │
└────┬─────┘ │
│ │
┌──┴───┐ │
▼ ▼ │
yes no │
│ │ │
│ ┌──────────────┐
│ │ 问题重写/改写 │ ◄── 优化 Query 后重新检索
│ └──────┬───────┘
│ │
│ ┌─────┘ (循环)
│ │
▼ ▼
┌──────────┐
│ 大模型生成 │ ◄── RAG 上下文 + LLM
└────┬─────┘
│
┌──┴──────┐
▼ │
yes no
│ │
▼ ▼
RAG回答 降级回答 ◄── 原始模型直接回答
项目结构
.
├── graph.py # LangGraph 工作流编排
├── router.py # LLM 链定义(路由/评分/重写/生成/降级)
├── RAGtool.py # 文档解析 → 分块 → 向量化 → 存储 → 检索
├── requirements.txt # 依赖清单
├── test.py # 测试脚本
├── .env # 环境变量(需自行创建)
└── pdf_file/ # 待索引的 PDF 文件
快速开始
1. 环境准备
git clone https://github.com/jyh20030112/ReflectionRAG.git
cd ReflectionRAG
pip install -r requirements.txt
2. 配置环境变量
创建 .env 文件:
API_KEY=sk-your-dashscope-key
EMBEDDING_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
EMBEDDING_MODEL=text-embedding-v3
CHAT_MODEL=qwen3-max
QDRANT_URL=https://your-cluster.qdrant.io
QDRANT_API_KEY=your-qdrant-key
QDRANT_COLLECTION=RAG_SEVER
TAVILY_API_KEY=tvly-your-tavily-key
3. 上传文档并写入向量库
from RAGtool import RAGTOOL
tool = RAGTOOL(file_path="./pdf_file/your-document.pdf")
result = tool.load_document()
print(result)
4. 启动问答
from graph import build_app
app = build_app()
result = app.invoke({"question": "大语言模型的底层机制是什么?"})
print(result["generation"])
核心模块
graph.py — LangGraph 工作流
| 节点 | 功能 |
|---|---|
route |
LLM 判断问题类型,决定走 RAG 还是 Web Search |
vectorestore |
调用 Qdrant 向量检索 |
web_search |
调用 Tavily 网络搜索 |
grade_documents |
LLM 逐一评分文档相关性 |
transform_query |
文档全不相关时重写 Query |
generation |
基于上下文调用大模型生成答案 |
router.py — LLM 链
| 链 | 用途 |
|---|---|
bulid_question_router() |
结构化输出路由判断 (vectorstore/web_search) |
build_retrieval_grader() |
结构化输出文档相关性评分 (yes/no) |
build_question_rewriter() |
查询重写优化 |
build_rag_chain() |
RAG 上下文增强生成 |
build_model_chain() |
降级时原始模型直接回答 |
RAGtool.py — 文档处理流水线
| 方法 | 功能 |
|---|---|
_enhanced_pdf_processing() |
MarkItDown PDF 增强解析 |
_split_paragraphs_with_headings() |
按 Markdown 标题分段 |
_chunk_paragraphs() |
Token 估算自适应分块 |
_embed_chunks() |
批量向量化 + Qdrant 分批写入 |
search() |
语义检索返回 Top-K 结果 |
技术栈
| 类别 | 技术 |
|---|---|
| 工作流编排 | LangGraph |
| LLM 框架 | LangChain |
| 大模型 | 通义千问 Qwen (DashScope API) |
| 向量数据库 | Qdrant Cloud |
| Embedding | DashScope text-embedding-v3 |
| 网络搜索 | Tavily Search API |
| 文档解析 | MarkItDown |
| 环境管理 | python-dotenv |
License
MIT
Metadata
Release files for simple_rag_reflect 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| simple_rag_reflect-0.1.1.tar.gz | 15.8 kB | Without provenance | Details for simple_rag_reflect-0.1.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| simple_rag_reflect-0.1.1-py3-none-any.whl | Python 3 | none | any | Details for simple_rag_reflect-0.1.1-py3-none-any.whl |
Total release size: 30.6 kB
Release files / simple_rag_reflect-0.1.1.tar.gz
| Download URL | simple_rag_reflect-0.1.1.tar.gz |
|---|---|
| Size | 15.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
bdc2f2dd87cd73f5a8cbf16d288fbee7048935f27ab68f0b7bbeb0167d1aa946
|
|
BLAKE2b-256 checksum How to use checksums |
4db0d9f5f5e3bf2987e9b164575e0e8e76040004ab470def58a6cd8db08b2e95
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.11.16 {"installer":{"name":"uv","version":"0.11.16","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
|
Release files / simple_rag_reflect-0.1.1-py3-none-any.whl
| Download URL | simple_rag_reflect-0.1.1-py3-none-any.whl |
|---|---|
| Size | 14.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
54dcd4a894de3831772be7508e5777b1064ae797785c502190e161f825227c1c
|
|
BLAKE2b-256 checksum How to use checksums |
6c7825e4ad82eb0cbde3d0b14535691096e41e64bc61fc9b9ff4a869a1ab4471
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
uv/0.11.16 {"installer":{"name":"uv","version":"0.11.16","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
|
Security for simple_rag_reflect 0.1.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.