Vector Embeddings & HNSW Similarity Search

// Vector Similarity Calculation & Embedding Search
import { cosineSimilarity } from 'ai-vector-math';

export interface VectorDocument {
  id: string;
  text: string;
  embedding: number[]; // 1536-dim vector
}

export class VectorStore {
  private docs: VectorDocument[] = [];

  add(doc: VectorDocument) { this.docs.push(doc); }

  search(queryVector: number[], topK: number = 3): VectorDocument[] {
    return [...this.docs]
      .map(doc => ({
        doc,
        similarity: cosineSimilarity(queryVector, doc.embedding)
      }))
      .sort((a, b) => b.similarity - a.similarity)
      .slice(0, topK)
      .map(item => item.doc);
  }
}