Multimodal content analysis for .NET.

Multimodal

Multimodal processing

Analyze both text and images with unified APIs. Automatic OCR integration extracts content from scanned documents, screenshots, and photos.

Custom

Fine-tunable models

Every analysis engine includes CreateTrainingObject methods for custom fine-tuning. Generate LoRA adapters tailored to your domain and data.

• Define unlimited custom categories
• Multimodal: classify both text and images
• GetBestCategory for single result, GetTopCategories for ranked list
• Guidance property for custom classification instructions
• HandleUnknownCategory for out-of-scope inputs

var model = LM.LoadFromModelID("qwen3.5:4b");
var classifier = new Categorization(model);

// Define your categories
var categories = new[] {
    "Technical Support", "Billing Inquiry",
    "Feature Request", "General Feedback"
};

// Classify incoming text
int result = classifier.GetBestCategory(
    categories, "My payment didn't go through");
Console.WriteLine(categories[result]);
// Output: Billing Inquiry

Content classification page

• Built-in entity types: Person, Organization, Location, Date, Money, Percent
• Define custom entity types with EntityDefinition
• Occurrences property with character offsets for each mention
• Multimodal: extract entities from images with OCR
• Guidance property for domain-specific extraction rules

var model = LM.LoadFromModelID("qwen3.5:4b");
var ner = new NamedEntityRecognition(model);

// Define entity types to extract
ner.EntityDefinitions.Add(new EntityDefinition(NamedEntityType.Person));
ner.EntityDefinitions.Add(new EntityDefinition(NamedEntityType.Organization));
ner.EntityDefinitions.Add(new EntityDefinition(
    "Product", "Product names mentioned"));

string text = "...";

NER page

• Text embeddings with models like nomic-embed-text, bge-m3
• Image embeddings with nomic-embed-vision
• Batch processing with GetEmbeddingsAsync for multiple inputs
• GetCosineSimilarity for measuring semantic similarity
• Seamless integration with RAG engine and vector stores

var model = LM.LoadFromModelID("nomic-embed-text");
var embedder = new Embedder(model);

// Generate text embeddings
float[] queryVec = await embedder.GetEmbeddingsAsync(
    "machine learning algorithms");

// Batch embed multiple texts
var docs = new[] { "AI overview", "Neural networks" };
float[][] docVecs = await embedder.GetEmbeddingsAsync(docs);

// Calculate similarity
float sim = Embedder.GetCosineSimilarity(queryVec, docVecs[0]);

Embeddings page

SentimentAnalysis

Classify text sentiment as positive, negative, or neutral with multilingual support and fine-tuning capabilities.

View docs

EmotionDetection

Identify specific emotions: joy, anger, sadness, fear, surprise, and disgust with optional neutral support.

View docs

Categorization

Custom classification with unlimited categories. Supports text and image inputs with confidence scores.

View docs

KeywordExtraction

Extract key terms and phrases with configurable count and language targeting.

View docs

NamedEntityRecognition

Extract people, organizations, locations, dates, and custom entity types with occurrence tracking.

View docs

PiiExtraction

Detect PII including names, emails, SSNs, and custom identifiers for compliance workflows.

View docs

SarcasmDetection

Detect ironic and sarcastic tones that might mislead sentiment analysis.

View docs

Embedder

Generate semantic embeddings from text and images for search, clustering, and RAG applications.

View docs

Reranker

Rerank search results by semantic relevance for improved RAG accuracy.

View docs