Table of Contents

👉 Try the demo: https://github.com/LM-Kit/lm-kit-net-samples/tree/main/console_net/rag-and-knowledge/query-expansion/query_strategies_lab

RAG Query Strategies Lab for C# .NET RAG Pipelines


🎯 Purpose of the Demo

An interactive console app that compares three retrieval strategies on the same RAG index: Original (one-shot embedding), MultiQuery (LLM rewrites the query into N variants), and HypotheticalAnswer (HyDE: LLM drafts a fake answer, then embeds it). Bring your own passages or use the built-in 7-passage seed.

All processing runs on-device.


👥 Industry Target Audience

  • RAG / search engineers evaluating recall lift on real corpora.
  • Knowledge base teams: pick the cheapest strategy that recovers your hardest queries.
  • eDiscovery / legal: when "the same idea phrased differently" still has to match.
  • Internal search: handle paraphrased questions in support chatbots.

🚀 Problem Solved

Naive RAG drops recall the moment a user phrases the question differently from the corpus. MultiQuery and HyDE are the two cheapest, well-known recall fixes. The lab puts them side-by-side so the impact is visible immediately on your own data, with all retrieved partitions printed per strategy for inspection.


💻 Application Overview

Interactive menu (no command-line arguments) with four modes:

Mode What it does
Compare Type a question, see the retrieved partitions and answer under all three strategies.
Import Paste passages to add to the index.
ImportFile Load passages from a UTF-8 text file (one per line).
Reset Wipe the index and reseed the default 7-passage corpus.
Quit Exit.

Both the embedding and chat models load once at startup. The default corpus is preloaded.

✨ Key Features

  • RagChat.QueryGenerationMode: Original, MultiQuery, HypotheticalAnswer.
  • MultiQueryOptions.QueryVariantCount: number of LLM-rewritten variants.
  • HydeOptions.MaxCompletionTokens: size of the synthetic answer.
  • RagQueryResult.RetrievedPartitions: full per-strategy retrieval inspection.

🧠 Models

  • qwen3-embedding:0.6b for the bi-encoder.
  • qwen3.5:4b for query rewriting and HyDE drafting.

🛠️ Getting Started

📋 Prerequisites

  • .NET 8.0 or later

▶️ Running the Application

git clone https://github.com/LM-Kit/lm-kit-net-samples
cd lm-kit-net-samples/console_net/rag-and-knowledge/query-expansion/query_strategies_lab
dotnet run

Pick a mode from the menu.

📚 Related Content

Share