Hybrid Search

📍 Question Answering Pipeline: Query Rewriting → Embedding → Filtering → Retrieval → Re-ranking → Context Build

Why Hybrid Search?

Pure vector search excels at capturing semantic meaning — "cancel my subscription" matches "terminate membership" even though they share no words. However, it can miss exact terms like product names, error codes, or policy identifiers that a user types verbatim.

BM25 keyword search handles these cases perfectly but fails at semantic understanding. Hybrid search combines both, giving you the best of both worlds: semantic understanding plus precise keyword matching.

Configuration

Blend dense vector search with BM25 keyword search using a single method call:

.WithRag(rag => rag
    .UseHybridSearch(vectorWeight: 0.6f)  // 60% vector, 40% BM25
    .AddDocument("knowledge-base.txt")
)

vectorWeight ranges from 0.0 (pure BM25) to 1.0 (pure vector). A value around 0.5–0.7 works well in most cases.

When to Use What

Scenario	Recommended Weight
General Q&A with natural language	0.7–0.8 (more vector)
Technical docs with specific terms	0.4–0.5 (balanced)
Code or error-code lookup	0.2–0.3 (more BM25)

Example

var service = new OpenAIService(apiKey, http)
    .WithRag(rag => rag
        .UseHybridSearch(vectorWeight: 0.5f)
        .AddDocument("product-catalog.txt")
        .AddDocument("error-codes.txt")
    );

// "ERR-4012" is matched by BM25; semantic context is matched by vector
var answer = await service.GetCompletionAsync("How do I fix ERR-4012?");