👉 Try the demo: https://github.com/LM-Kit/lm-kit-net-samples/tree/main/console_net/rag/help_desk_knowledge_base

Help Desk Knowledge Base for C# .NET Applications

🎯 Purpose of the Sample

The Help Desk Knowledge Base demo shows how to build a production-grade RAG system with LM-Kit.NET that persists its index to disk, supports incremental article management, and scopes queries to specific categories using DataFilter.

This is the demo to study when you need to build a knowledge base that survives restarts, grows over time, and serves different departments or topics.

👥 Industry Target Audience

This demo is particularly useful for developers and organizations working on:

Customer support: build a knowledge base of FAQs and support articles that agents (human or AI) query to resolve tickets
Internal documentation portals: maintain a searchable wiki of company policies, procedures, and technical documentation
Multi-department help desks: scope queries to specific departments (HR, IT, Finance) using category filters
SaaS product documentation: embed a searchable knowledge base in your product for self-service support
Compliance and legal: maintain a persistent, auditable knowledge base of regulations and policies

🚀 Problem Solved

Most RAG tutorials show a "build once, query, throw away" pattern. Real applications need to:

Persist the index so it survives restarts without re-embedding (which is slow and costly)
Add and remove content incrementally as documentation evolves
Scope queries so users searching billing issues do not get results from technical troubleshooting
Show sources so users can verify answers against the original articles

This demo solves all of these problems with a complete, working implementation.

💻 Sample Application Description

The Help Desk Knowledge Base demo is a console application that:

Loads a chat model and an embedding model
Creates or reloads a persistent, file-backed knowledge base
Seeds 20 fictional support articles across 5 categories on first run
Provides an interactive help desk assistant with article management commands

✨ Key Features

Persistent storage: file-backed DataSource that reloads instantly without re-embedding
Incremental management: add and remove articles at runtime; changes persist automatically
Category-scoped search: DataFilter restricts queries to specific categories
Markdown-aware chunking: MarkdownChunking respects heading boundaries for cleaner chunks
Answer generation: two-step pipeline using FindMatchingPartitions then QueryPartitions
Source attribution: shows which articles contributed to each answer, with relevance scores

🏗️ Architecture

    ┌─────────────────────────────────────────────────────────┐
    │                   help_desk_kb.dat                      │
    │              (persistent file storage)                  │
    │                                                         │
    │  Account Management/  Billing/  Technical Support/ ...  │
    │    Password Reset       Plans     Connectivity          │
    │    Account Recovery     Payments  Error Codes           │
    │    Profile Settings     Refunds   API Rate Limits       │
    │    ...                  ...       ...                   │
    └──────────────────────────┬──────────────────────────────┘
                               │
                    LoadFromFile / CreateFileDataSource
                               │
                    ┌──────────▼──────────┐
                    │     RagEngine       │
                    │  + DataFilter       │◄── /scope Billing
                    └──────────┬──────────┘
                               │
                    FindMatchingPartitions
                               │
                    ┌──────────▼──────────┐
                    │  Top-5 Passages     │
                    │  + source info      │──► Source Attribution
                    └──────────┬──────────┘
                               │
                      QueryPartitions
                               │
                    ┌──────────▼──────────┐
                    │  SingleTurnConv     │
                    │  (chat model)       │──► Grounded Answer
                    └─────────────────────┘

🛠️ Getting Started

📋 Prerequisites

.NET 8.0 or later
Minimum 6 GB VRAM (Qwen-3 8B) or 4 GB VRAM (Gemma 3 4B)

📥 Download the Project

.NET Console Demo

▶️ Running the Application

Clone the repository:

git clone https://github.com/LM-Kit/lm-kit-net-samples

Navigate to the project directory:

cd lm-kit-net-samples/console_net/rag/help_desk_knowledge_base

Build and run the application:
```
dotnet build
dotnet run
```
First run: the knowledge base is created and seeded with 20 sample articles. Subsequent runs reload the persisted index instantly.

💡 Example Usage

Ask a question: type a support question to get a grounded answer with source attribution
Scope to a category: use /scope Billing to restrict queries, then /scope all to remove the filter
Add an article: use /add Security "API Key Best Practices" and enter content interactively
Remove outdated content: use /remove "Outdated Article" to remove articles
Browse the knowledge base: use /categories for an overview or /list Technical Support for details
Restart the demo: the knowledge base reloads from disk without re-embedding

🔧 Troubleshooting

Issue	Solution
No relevant articles found	Remove the scope filter with `/scope all` or lower the search threshold
First run is slow	Embedding 20 articles takes a few seconds; subsequent runs are instant
Article not added	Check that the section identifier (Category/Title) does not already exist

🚀 Extend the Demo

Load from files: use /addfile to import Markdown files, or modify the code to scan a directory
Add hybrid search: set ragEngine.RetrievalStrategy = new HybridRetrievalStrategy() for BM25 + vector fusion
Multi-turn conversation: replace SingleTurnConversation with RagChat for follow-up questions
Connect to Qdrant: replace the file-backed DataSource with a Qdrant-backed store for scalability

📚 Additional Resources

LM-Kit.NET RAG Documentation
Conversational RAG Demo: multi-turn RAG with query generation modes
Retrieval Quality Tuning Demo: compare Vector, BM25, and Hybrid strategies
Single-Turn RAG with Qdrant Demo: enterprise RAG with external vector database

Table of Contents