import { prisma } from '../db.js';

// Simulated embedding generation for Phase 14 demo
// In production, this would call OpenAI text-embedding-3-small or similar
export async function generateEmbedding(text: string): Promise<number[]> {
  // Deterministic-ish random for demo purposes based on text length
  const vector = new Array(768).fill(0).map((_, i) => Math.sin(text.length * i));
  // Normalize vector
  const magnitude = Math.sqrt(vector.reduce((sum, val) => sum + val * val, 0));
  return vector.map(val => val / magnitude);
}

export async function storeArticleEmbedding(articleId: string, text: string) {
  const vector = await generateEmbedding(text);
  
  // Format for pgvector: '[0.1, 0.2, ...]'
  const vectorString = `[${vector.join(',')}]`;

  await prisma.$executeRaw`
    INSERT INTO "ArticleEmbedding" ("id", "articleId", "embedding", "updatedAt")
    VALUES (gen_random_uuid(), ${articleId}, ${vectorString}::vector, NOW())
    ON CONFLICT ("articleId") DO UPDATE SET
      "embedding" = EXCLUDED."embedding",
      "updatedAt" = NOW();
  `;
}

export async function semanticSearch(query: string, limit = 5) {
  const queryVector = await generateEmbedding(query);
  const vectorString = `[${queryVector.join(',')}]`;

  // Cosine distance operator <=> in pgvector
  const results = await prisma.$queryRaw<Array<{ id: string; articleId: string; distance: number; title: string; slug: string }>>`
    SELECT 
      e."id", 
      e."articleId", 
      (e."embedding" <=> ${vectorString}::vector) as distance,
      a."title",
      a."slug"
    FROM "ArticleEmbedding" e
    JOIN "Article" a ON e."articleId" = a."id"
    ORDER BY distance ASC
    LIMIT ${limit};
  `;

  return results;
}