Vektor-Datenbanken

import { VectorStore } from "std:db";
import { ai } from "std:ai";

// Pinecone verbinden
let vectorDB = VectorStore.connect("pinecone", {
    index: "my-documents",
    apiKey: env.PINECONE_KEY,
});

// Dokument speichern mit Embedding
@POST("/api/documents")
fn createDocument(title: string, content: string): Document {
    // Embedding generieren
    let embedding = ai.embeddings(content, "text-embedding-3-small");
    
    // In Vektor-DB speichern
    vectorDB.upsert({
        id: generateId(),
        values: embedding,
        metadata: { title, content },
    });
    
    return Document { title, content };
}

@POST("/api/search")
fn searchDocuments(query: string, topK: number): List<SearchResult> {
    // Query zu Embedding umwandeln
    let queryVector = ai.embeddings(query, "text-embedding-3-small");
    
    // Ähnliche Dokumente finden
    let results = vectorDB.query({
        vector: queryVector,
        topK: topK,
        includeMetadata: true,
    });
    
    return results.map(r => SearchResult {
        title: r.metadata.title,
        content: r.metadata.content,
        score: r.score,
    });
}

// Beispiel-Aufruf:
// POST /api/search { query: "Wie funktioniert KI?", topK: 5 }
// → Findet Dokumente über Machine Learning, neuronale Netze, etc.