# LDK_QA

**Repository Path**: cris721/LDK_QA

## Basic Information

- **Project Name**: LDK_QA
- **Description**: 基于大模型的垂域知识库问答系统平台
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 3
- **Forks**: 0
- **Created**: 2026-04-08
- **Last Updated**: 2026-04-29

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Alfred

A domain-specific knowledge base QA system powered by LLM + RAG + multi-session web chat.

---

## 1. Project Overview

Alfred is a vertical-domain QA platform that combines general LLM capabilities with domain knowledge retrieval to deliver practical, production-oriented intelligent Q&A experiences.

The project currently focuses on local development and uses a frontend-backend separated architecture:

- Backend uses Django + Django Ninja to provide REST APIs and streaming endpoints.
- Frontend uses Vue 3 + Vite for a chat-style user interface.
- SQLite stores user/account/session/config/file metadata.
- ChromaDB is used for vector retrieval in expert-mode RAG.

---

## 2. Core Features

### 2.1 Account and Authentication

- User registration, login, and bearer-token authentication.
- User profile retrieval and password update.
- Token lifecycle management with expiry.

### 2.2 Chat and Session Management

- Supports standard chat (`/api/chat`) and streaming chat (`/api/chat/stream`).
- Supports creating, listing, and deleting multiple sessions.
- Automatically persists conversation history and supports per-session history clearing.

### 2.3 Dual Answering Modes (daily / expert)

- `daily`: general conversation mode for regular Q&A.
- `expert`: RAG-enabled mode prioritizing knowledge evidence.
- Automatically falls back to direct model answer with a no-hit notice when expert retrieval misses.

### 2.4 User File Knowledge Base

- Supports uploading `txt/csv/md/log/json/pdf` files.
- Automatically extracts text and vectorizes uploaded content.
- Supports per-session file listing/deletion with strict user/file retrieval isolation.

### 2.5 Model and Provider Configuration

- Supports local `Ollama`.
- Supports online providers (OpenAI-compatible APIs and Anthropic).
- Supports dynamic model list fetching and user-level saved config.

### 2.6 Frontend Experience

- ChatGPT-style chat page and session sidebar.
- Complete flow for login/register/settings.
- Supports light/dark themes, multi-language UI, and account management.

---

## 3. Tech Stack

### 3.1 Backend

- Python 3.x
- Django 4.2+
- Django Ninja
- django-cors-headers
- requests

### 3.2 RAG and Data Processing

- ChromaDB (vector storage)
- rank-bm25 (lexical reranking)
- pypdf (PDF text extraction)

### 3.3 Frontend

- Vue 3
- Vite
- Pinia
- Vue Router
- Axios

### 3.4 Storage and Deployment Shape

- SQLite (structured business data)
- Chroma persistent path: `django_backend/data/rag_chroma`
- Currently optimized for local development deployment

---

## 4. Installation Guide

> Steps below target Linux/macOS. For Windows, WSL is recommended.

### 4.1 Clone the Repository

```bash
git clone <your-repo-url>
cd LDK_QA
```

### 4.2 Start Backend (Django)

```bash
cd django_backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python manage.py migrate
python manage.py runserver 0.0.0.0:8081
```

### 4.3 Start Frontend (Vue)

Open another terminal:

```bash
cd vue_frontend
npm install
npm run dev
```

Default frontend URL: `http://localhost:8082`  
Vite proxies `/api` requests to `http://localhost:8081`.

### 4.4 Prepare Model Runtime (Recommended)

If you use local models, install and run Ollama first, then pull models (example):

```bash
ollama pull qwen2.5:0.5b
ollama pull nomic-embed-text:latest
```

---

## 5. Usage Guide

### 5.1 First-Time Flow

1. Open the frontend page and register an account.
2. Log in and enter the chat page.
3. Select mode (`daily`/`expert`) and model in settings.
4. In expert mode, upload files to improve evidence quality.
5. Send questions and review standard or streaming responses.

### 5.2 Recommended Paths

- Daily Q&A: use `daily` mode directly.
- Business/domain Q&A: switch to `expert`, upload domain material, then ask.
- Follow-up discussions: stay in the same session to keep context.
- Context reset: clear history for one session or create a new session.

### 5.3 Optional Ops Command (Domain KB Ingestion)

Run in `django_backend`:

```bash
python manage.py bootstrap_rag --domain_id default
```

Force rebuild for a domain:

```bash
python manage.py bootstrap_rag --domain_id default --force
```

---

## 6. Suggested Team Split (Based on Current Features)

To keep iteration efficient, organize work by business flow rather than isolated technologies:

### 6.1 Product and Requirements (1 person)

- Prioritize scenarios and boundaries for daily vs expert mode.
- Maintain feature backlog, acceptance criteria, and release rhythm.
- Define answer quality standards and prompt expectations.

### 6.2 Frontend Module (1-2 people)

- Chat page (sessions, message flow, streaming rendering).
- Settings page (model config, theme, language, account profile).
- Upload/file UX (progress, status hints, error handling).

### 6.3 Backend API and Business Logic (1-2 people)

- APIs for auth, sessions, history, files, and model config.
- Auth enforcement, parameter validation, status codes, stability.
- Streaming endpoint consistency and performance tuning.

### 6.4 RAG and Model Capability (1-2 people)

- Vector ingestion/retrieval/reranking and hit-quality optimization.
- Expert-mode prompt and evidence citation quality.
- No-hit fallback strategy and retrieval observability.

### 6.5 QA and Operations (1 person)

- Regression testing (login/chat/upload/streaming).
- DB migrations, startup scripts, env var conventions.
- Pre-release checks (integration, performance, logs/monitoring).

---

## Project Structure (Simplified)

```text
LDK_QA/
├── django_backend/        # Django backend + RAG logic
│   ├── ai_api/            # Core APIs, models, services
│   └── manage.py
├── vue_frontend/          # Vue frontend
└── README.md
```