const results = await collection.query({
  vector: [0.1, 0.2, 0.3, ...],  // your query embedding
  topK: 10
});

results.forEach(result => {
  console.log(`${result.id}: score=${result.score.toFixed(4)}`);
  console.log(`  ${result.metadata.text}`);
});

interface QueryResult {
  id: string; // vector ID
  score: number; // similarity score (higher = more similar)
  metadata: Record<string, any>; // stored metadata
  values?: number[]; // raw vector values (only if includeValues: true)
}

// Only return results with cosine similarity > 0.75
const results = await collection.query({
  vector: queryEmbedding,
  topK: 10,
  minScore: 0.75,
});

// Good thresholds for cosine similarity:
// > 0.9  - very similar, almost paraphrases
// > 0.8  - similar topic and content
// > 0.7  - related content
// > 0.5  - loosely related
// < 0.5  - probably not relevant

import { SolVec } from "@veclabs/solvec";

async function semanticSearch(userQuery: string) {
  const sv = new SolVec({ network: "devnet" });
  const collection = sv.collection("knowledge-base", { dimensions: 1536 });

  // 1. Embed the query
  const queryEmbedding = await embed(userQuery);

  // 2. Search
  const results = await collection.query({
    vector: queryEmbedding,
    topK: 5,
    minScore: 0.7,
  });

  // 3. Handle no results
  if (results.length === 0) {
    return { answer: "No relevant information found.", sources: [] };
  }

  // 4. Build LLM context from results
  const context = results.map((r) => r.metadata.text).join("\n\n");

  // 5. Generate answer
  const answer = await callLLM(
    `Answer using this context:\n\n${context}\n\nQuestion: ${userQuery}`,
  );

  return {
    answer,
    sources: results.map((r) => ({
      id: r.id,
      score: r.score,
      source: r.metadata.source,
      text: r.metadata.text.slice(0, 200) + "...",
    })),
  };
}

try {
  const results = await collection.query({
    vector: queryEmbedding,
    topK: 10,
  });
} catch (error) {
  if (error.code === "DIMENSION_MISMATCH") {
    // Query vector has wrong number of dimensions
    console.error("Query vector dimensions do not match collection");
  } else if (error.code === "EMPTY_INDEX") {
    // Collection has no vectors yet
    console.log("Collection is empty - no results to return");
  } else {
    throw error;
  }
}