DMMRSuggestionEngine 0.1.7

DMMR Suggestion Engine

A high-performance fuzzy search and hybrid suggestion engine for .NET.

DMMR Suggestion Engine combines BK-Tree indexing, N-Gram search, hybrid OpenSearch integration, similarity metrics, and an extensible filter pipeline to deliver fast, relevant, and customizable search suggestions.

Designed for:

• Product Search
• Autocomplete
• Search Suggestions
• Typo Correction
• Recommendation Systems
• Hybrid Search
• AI Search Pipelines

Installation

dotnet add package DMMRSuggestionEngine

Features

• ✅ BK-Tree fuzzy search
• ✅ Character N-Gram indexing
• ✅ Typo-tolerant search
• ✅ Weighted ranking
• ✅ Built-in LRU cache
• ✅ Generic type-safe API
• ✅ Hybrid OpenSearch search
• ✅ ReadOnlySpan optimized algorithms
• ✅ Low memory allocations
• ✅ Extensible filter pipeline
• ✅ Async APIs
• ✅ .NET 9

Quick Start

using DMMRSuggestionEngine;

var engine = new DMMRSuggestionEngine<(int Id, string Name, float Weight)>();

engine.LoadData(
    new[]
    {
        (1, "iPhone 16 Pro Max", 100f),
        (2, "Samsung Galaxy S25 Ultra", 80f),
        (3, "Motorola Edge 60", 50f)
    },
    x => x.Name,
    x => x.Weight);

var results = engine.Suggest("iphon");

N-Gram Search

The built-in N-Gram index complements BK-Tree search by handling queries such as:

Configuration:

engine.NgramConfig.Enabled = true;
engine.NgramConfig.N = 3;
engine.NgramConfig.MinScore = 0.2f;

Hybrid OpenSearch

using DMMRSuggestionEngine.OpenSearch;
using OpenSearch.Client;

var client = new OpenSearchClient(
    new ConnectionSettings(
        new Uri("http://localhost:9200")));

var hybrid = new DMMROHybridSearchService(
    client,
    "products");

var results = await hybrid.HybridSearchAsync(
    "iphone",
    embedding);

Features:

• Local + Distributed Search
• Async Search
• CancellationToken support
• Vector Search
• OpenSearch Integration

Filter Pipeline (v0.1.7)

Hybrid search supports an extensible post-processing pipeline.

Multiple filters can be chained to refine search results.

using DMMRSuggestionEngine.OpenSearch.Filters;

service.AddFilter(new MinWeightFilter(50));

service.AddFilter(new CosineSimilarityFilter(0.40f));

service.AddFilter(new AbortIfEmptyFilter());

var results = await service.HybridSearchAsync(
    "laptop",
    embedding);

Built-in filters:

Create Your Own Filters

The pipeline is fully extensible.

Simply implement IPostFilter.

public class MyFilter : IPostFilter
{
    public List<SuggestionDocument> Apply(
        List<SuggestionDocument> results,
        SearchContext context,
        out bool continueProcessing)
    {
        continueProcessing = true;

        // Custom business rules

        return results;
    }
}

Register it:

service.AddFilter(new MyFilter());

This makes it easy to implement:

• Inventory filtering
• Category filtering
• Region restrictions
• Customer segmentation
• Logging
• Metrics
• Business rules
• Deduplication
• Any custom post-processing

Performance

Designed for:

• Low latency
• Low allocations
• High throughput
• Large datasets

Typical search latency:

Main Components

• BK-Tree
• Character N-Gram Index
• Levenshtein Distance
• Cosine Similarity
• Pearson Correlation
• Jaccard Similarity
• Euclidean Distance
• Manhattan Distance
• LRU Cache
• OpenSearch Hybrid Search
• Filter Pipeline

Why DMMR?

Unlike traditional fuzzy search libraries, DMMR combines multiple search strategies into a single engine:

• Fast typo correction
• Partial phrase matching
• Hybrid local/distributed search
• Vector similarity
• Extensible post-processing pipeline
• Performance-oriented implementation
• Minimal allocations
• Generic API

License

MIT

Author

Alessandro Silveira

Software Architect • DevOps Tech Leader • Performance-Oriented Systems Engineer

GitHub:

https://github.com/rootmese