This document provides a curated set of example questions to demonstrate how DataMind's Agent automatically dispatches different modules to answer questions. Each question notes which module is expected to run, so you can verify the system behaves as intended.

How to use: start the Web UI (python server.py) or the terminal (python main.py), then enter the questions below in order.

1. RAG knowledge base retrieval

The RAG module retrieves semantically relevant content from documents under the profile directory.

#	Example question	Expected behavior
1.1	DataMind 支持哪些文档格式？	Retrieves sample.txt; returns PDF/TXT/MD/DOCX, etc.
1.2	这个项目使用了什么向量数据库？	Retrieves Chroma-related description
1.3	RAG 检索的核心原理是什么？	Retrieves description of vectorization + cosine similarity
1.4	LlamaIndex 提供了哪些功能？	Retrieves data connectors, index structures, query interfaces

2. Multimodal RAG retrieval

Multimodal RAG can extract information from images and use it for retrieval-based Q&A. Switch to the multimodal demo profile first:

DATA_PROFILE=mm_demo IMAGE_EMBEDDING_MODE=vlm_describe python server.py

The built-in mm_demo profile contains mixed text + image data (data/profiles/mm_demo/), with three images showing a system architecture diagram, a retrieval strategy comparison bar chart, and a knowledge graph visualization.

#	Example question	Expected behavior
2.1	系统架构有哪几层？每层包含什么组件？	VLM extracts architecture info from arch.png; answer includes Data Layer / Service Layer / API Gateway, etc.
2.2	哪种检索策略的召回率最高？具体是多少？	VLM reads bar chart data from chart.png; answer includes Hybrid 91%, etc.
2.3	知识图谱中有哪些实体和关系？	VLM extracts graph structure from graph.png; answer includes entity/edge counts

Note: On the first run in multimodal mode, the system calls a VLM API (e.g. GPT-4o) to generate text descriptions for images. Descriptions are cached in the index afterwards.

3. GraphRAG graph retrieval

The GraphRAG module performs multi-hop reasoning over entities and relations in the knowledge graph.

#	Example question	Expected behavior
3.1	DataMind 基于什么框架？	Graph reasoning: DataMind → based on → LlamaIndex
3.2	DataMind 包含哪些模块？	Graph reasoning: DataMind → contains modules → RAG/GraphRAG/NL2SQL
3.3	RAG 向量检索使用了什么技术？	Graph reasoning: RAG vector retrieval → uses technology → Chroma
3.4	LlamaIndex 和 Python 是什么关系？	Graph reasoning: LlamaIndex → is → Python framework

4. Database queries

The Database module turns natural language into SQL and runs it against SQLite.

The demo database currently includes:

• employees table: 8 employees (张三, 李四, 王五, 赵六, 孙七, 周八, 吴九, 郑十)
• projects table: 4 projects (RAG智能助手, 数据分析平台, 移动端App, 品牌推广)

#	Example question	Expected behavior
4.1	工程部有几个人？	SQL: SELECT COUNT(*) ... WHERE department='工程部' → 4
4.2	谁的工资最高？	SQL: ORDER BY salary DESC LIMIT 1 → 孙七 45000
4.3	北京的员工有哪些？	SQL: WHERE city='北京' → 张三, 王五, 孙七, 郑十
4.4	预算超过 20 万的项目有哪些？	SQL: WHERE budget > 200000 → RAG智能助手, 数据分析平台
4.5	RAG 智能助手项目的负责人是谁？	SQL: JOIN employees and projects → 孙七
4.6	各部门的平均工资是多少？	SQL: GROUP BY department + AVG(salary)

5. Skills — knowledge skills

The Skills knowledge module retrieves procedures and best practices from Markdown files under data/skills/.

#	Example question	Expected behavior
5.1	数据库备份的最佳策略是什么？	Retrieves SOP; returns full + incremental backup strategy
5.2	数据库慢查询怎么排查？	Retrieves SOP; returns slow-query analysis steps
5.3	代码审查应该重点关注什么？	Retrieves review guide; correctness / quality / security / performance
5.4	Code Review 的反馈应该怎么写？	Retrieves review guide; MUST / SHOULD / NICE tiers
5.5	数据库故障排查的步骤？	Retrieves SOP troubleshooting checklist

6. Skills — tool skills

Tool skills are Python functions the Agent calls automatically for precise results.

#	Example question	Expected behavior
6.1	现在几点了？	Calls get_current_time → current date and time
6.2	计算 sqrt(144) + 3^4	Calls calculator → 12 + 81 = 93
6.3	100 公里等于多少英里？	Calls unit_convert → 62.14 miles
6.4	25 摄氏度等于多少华氏度？	Calls unit_convert → 77°F

7. Memory — conversation memory

The Memory module lets the Agent retain conversation context. Use a short sequence of questions to verify it.

#	Step	Expected behavior
7.1	First ask: 工程部有几个人？	Normal answer: 4
7.2	Then ask: 他们分别是谁？	Agent recalls the "工程部" context; returns 张三, 李四, 孙七, 吴九
7.3	Then ask: 其中谁的工资最高？	Agent uses context; answers: 孙七 45000
7.4	Ask: 我刚才问了什么？	Agent summarizes prior turns from memory

8. Multi-module orchestration

These prompts may cause the Agent to invoke multiple tools at once, showing intelligent dispatch.

#	Example question	Expected modules
8.1	DataMind 用了什么技术栈？各模块分别用了什么？	RAG + GraphRAG
8.2	工资最高的员工负责的是哪个项目？这个项目预算多少，折合多少美元？	Database + calculator
8.3	数据库出了性能问题，应该怎么排查？先帮我看看当前数据库有哪些表	Skills + Database
8.4	今天是几号？帮我算一下如果按全量备份策略，30天前的备份应该从哪天开始保留？	get_current_time + calculator + Skills

9. Casual chat

For these, the Agent answers with the LLM directly and does not need tools.

#	Example question	Expected behavior
9.1	你好！	Greeting
9.2	你能做什么？	Describes capabilities
9.3	给我讲个笑话	Generates content directly

Recommended demo flow

For a full walkthrough, use this order:

1. RAG: ask DataMind 支持哪些文档格式？
2. Multimodal RAG: switch to mm_demo profile, ask 系统架构有哪几层？
3. GraphRAG: ask DataMind 包含哪些模块？
4. Database: ask 谁的工资最高？
5. Skills: ask 数据库备份的最佳策略是什么？
6. Tool: ask 现在几点了？
7. Memory: ask 工程部有几个人？ then 他们分别是谁？
8. Orchestration: ask 工资最高的员工负责哪个项目？预算折合多少美元？

Tip: in the Web UI, open the right-hand panel to watch module status change as you ask questions.