AI knowledge base and intelligent Q&A

# 🤖 AI Knowledge Base and Intelligent Q&A

@[frag](id=7 title=Pro Edition Guide - demo category)

## 1. Overview

MrDoc includes a complete built-in AI RAG (Retrieval-Augmented Generation) engine, which can directly implement:

* Document vectorization
* Vector retrieval
* AI Q&A
* AI-assisted writing
* Document semantic search
* AI content generation

The built-in AI engine operates using a "RAG + Large Language Model" approach. Its core capabilities include:

* Document chunking
* Embedding vectorization
* Vector retrieval
* Rerank
* Large Language Model generation (LLM)

At the same time, MrDoc is not bound to any specific AI service provider. All AI capabilities can be implemented through independently configured large model interfaces, including:

* Text generation models
* Embedding models
* Rerank models

On this basis, MrDoc also supports integration with Dify to extend more complex AI workflows and application orchestration capabilities.

This document mainly introduces:

* The working mechanism of MrDoc's built-in AI engine;
* The integration method of Dify;
* The role of independent AI interfaces;
* The overall capabilities after AI integration.

---

## 2. AI Integration Options

MrDoc currently supports two AI integration options:

| Option | Description |
| --------- | ------------------ |
| Built-in AI Engine | Uses MrDoc's built-in RAG capabilities |
| Dify Integration | Uses Dify as the AI platform |

The two options:

* Have different functional positioning;
* Have different configuration methods;
* Have different operational complexity;

In general:

> Choosing one of the two options is sufficient.

---

### Option 1: Built-in AI Engine

The built-in AI engine is:

> A complete RAG knowledge base solution implemented by MrDoc itself.

The overall process is as follows:

```text
Document content
   ↓
Document chunking (Chunk)
   ↓
Embedding vectorization
   ↓
Vector storage
   ↓
User question
   ↓
Vector retrieval
   ↓
Rerank (optional)
   ↓
LLM generates answer
```

Its characteristics:

* Ready to use out of the box;
* Simple configuration;
* Low operational cost;
* More suitable for private deployment;
* More suitable for document knowledge base scenarios.

Applicable to:

* AI knowledge base
* AI knowledge Q&A
* Document semantic search
* AI-assisted writing
* Enterprise internal knowledge base

---

### Option 2: Dify Integration

Dify is a third-party AI platform.

MrDoc supports syncing documents to the Dify knowledge base and calling AI interfaces provided by Dify.

Dify is more oriented towards:

* AI workflow platform
* AI Agent platform
* Multi-model unified management platform
* AI application orchestration platform

Applicable to:

* Enterprise AI workflows
* AI Agent
* Multi-step AI automation
* Complex AI applications

Compared to the built-in AI engine:

* Higher operational complexity;
* Higher deployment cost;
* More suitable for complex AI scenarios.

---

## 3. MrDoc Built-in AI Engine

### Built-in AI Engine Architecture

The MrDoc built-in AI engine is essentially a complete RAG knowledge base solution.

The overall process is as follows:

```text
Document content
   ↓
Document chunking (Chunk)
   ↓
Embedding vectorization
   ↓
Vector storage
   ↓
User question
   ↓
Vector retrieval
   ↓
Rerank (optional)
   ↓
LLM generates answer
```

---

### Components of the Built-in AI Engine

#### (1) Embedding Model

Used to convert document content into vector data for semantic retrieval.

Mainly used for:

* Document vectorization
* User question vectorization
* Similarity retrieval

---

#### (2) Vector Database

Used to store document vector data.

By default:

* MrDoc can use a built-in storage solution;
* No additional vector database deployment is required.

It also supports configuring an external Qdrant vector database to achieve:

* Higher performance;
* Larger-scale vector storage;
* Stronger scalability.

---

#### (3) Rerank Model (Optional)

Used to perform secondary sorting on vector retrieval results to improve retrieval accuracy.

Applicable to:

* Large-scale knowledge bases;
* Scenarios with many similar contents;
* Scenarios with high requirements for Q&A accuracy.

---

#### (4) LLM Large Language Model

Used to generate the final answer content.

Can be used for:

* AI Q&A
* AI-assisted writing
* Document summarization
* Content polishing
* AI translation

---

## 4. Dify Integration Architecture

In addition to the built-in AI engine, MrDoc also supports integration with Dify.

Dify is more oriented towards:

* AI workflow platform
* AI application orchestration platform
* Multi-model unified management platform

Applicable to:

* Complex AI workflows;
* Enterprise-level AI applications;
* Multi-model unified scheduling;
* AI Agent scenarios.

---

### (1) Sync Documents to Dify Knowledge Base

MrDoc supports syncing documents to the Dify knowledge base for Dify to perform:

* Document chunking
* Vectorization
* AI Q&A

The synced content includes:

* Document title
* Document body

Supports:

* Manual sync
* Automatic sync

---

### (2) Call Dify AI Interfaces

MrDoc can directly call interfaces provided by Dify, including:

* Chat / Completion interfaces
* Knowledge base retrieval interfaces
* Dify App conversation interfaces

Used to implement:

* AI chat
* AI Q&A
* AI workflows
* Enterprise AI assistant

---

## 5. Independent AI Large Model Interfaces

Whether it is:

* Built-in AI engine
* AI-assisted writing
* Dify integration

Essentially, all require calling external AI large model interfaces.

Therefore, MrDoc supports separate configuration of:

| Type | Purpose |
| ------------ | ---------- |
| Text generation model | AI conversation, content generation |
| Embedding model | Document vectorization |
| Rerank model | Retrieval result reranking |

Supports integration with:

* OpenAI interface specifications
* Locally deployed models
* Third-party AI service providers
* OpenAI-compatible interfaces

Including but not limited to:

* OpenAI
* DeepSeek
* Qwen
* Gemini
* Claude
* Ollama

---

## 6. AI Knowledge Base Q&A Workflow

AI knowledge base Q&A uses a typical RAG (Retrieval-Augmented Generation) process.

The overall workflow is as follows:

```mermaid
flowchart TD

A[User question] --> B[Question vectorization]
B --> C[Vector database retrieval]
C --> D[Recall relevant chunks]
D --> E[Rerank]
E --> F[Concatenate context]
F --> G[Submit to LLM]
G --> H[Generate final answer]
```

---

### Workflow Description

#### (1) User Question

The user inputs a natural language question.

For example:

```text
How to deploy Qdrant?
```

---

#### (2) Question Vectorization

The system calls the Embedding model to convert the user question into a vector.

Used for semantic retrieval.

---

#### (3) Vector Retrieval

The system searches the vector database for document chunks most similar to the question.

---

#### (4) Rerank

If the Rerank model is enabled:

The system re-sorts the recalled results to improve accuracy.

---

#### (5) LLM Generates Answer

The system submits the retrieval results as context to the large language model.

The LLM ultimately generates a natural language answer.

---

## 7. AI Capabilities After Integration

After completing AI integration, MrDoc can achieve the following capabilities.

---

### AI Knowledge Base

* Document semantic search
* AI knowledge Q&A
* RAG retrieval-augmented generation
* Enterprise knowledge base

---

### AI Writing

* AI document creation
* AI content continuation
* AI polishing and optimization
* AI translation

---

### AI Reading and Analysis

* Document summarization
* Content extraction
* Intelligent analysis
* AI-assisted reading

---

### AI Conversation

* Website AI assistant
* Enterprise knowledge bot
* AI chat assistant

---

## 8. Prerequisites for Using AI

Before using AI features, the following configurations must be completed:

| Configuration Item | Required |
| ------------ | ---- |
| AI engine configuration | Yes |
| Text generation model | Yes |
| Embedding model | Yes |
| Document vectorization | Yes |

Optional configurations:

| Configuration Item | Required |
| ------------ | ---- |
| Rerank model | No |
| Qdrant vector database | No |

It is recommended to first read:

* Dify Framework Configuration
* Qdrant Deployment
* AI Document Creation

---

## 9. Recommended Usage Suggestions

| Usage Requirement | Recommended Option |
| ----------- | -------------------------- |
| Quickly enable AI | Built-in AI engine |
| Low operational complexity | Built-in AI engine |
| Private deployment knowledge base | Built-in AI engine |
| AI document writing | Independent AI interface |
| Enterprise-level AI workflows | Dify |
| Multi-model unified management | Dify |
| AI Agent scenarios | Dify |
| Large-scale knowledge base | Built-in AI engine + Qdrant + Rerank |

---

## 10. FAQ

### AI Answer Is Empty

Possible reasons:

* Documents not vectorized;
* AI sync not enabled;
* Embedding configuration error;
* Vector database abnormality;
* No relevant content retrieved.

---

### AI Answer Is Inaccurate

Possible reasons:

* Insufficient document content;
* Disorganized document structure;
* Low retrieval hit rate;
* Rerank not enabled;
* Insufficient large model capability.

---

### AI Cannot Understand Professional Terminology

Suggestions:

* Add terminology explanations in documents;
* Add FAQ;
* Add example content;
* Use industry-specific models.

---

## 11. Summary

MrDoc's AI capabilities are essentially:

> A deep integration of "document system + RAG + large language model."

Through the combination of AI and knowledge base, MrDoc is not only a document system but can also become:

* Enterprise AI assistant
* Privatized knowledge bot
* Intelligent retrieval platform
* AI-driven knowledge hub

In private deployment scenarios, it can effectively solve:

* Documents difficult to retrieve
* Knowledge difficult to accumulate
* AI unable to combine with internal data
* Data unable to leave the intranet

and other issues.