Contextual Ad Engine

const res = await tl_client.generate.text(videoId, {
  prompt: `Analyze each scene segment and return JSON with:
  - scene_context, sentiment, tone, environment
  - ad_suitability: { suitable_categories, unsuitable_categories }
  - brand_safety: { garm_flags, risk_level }
  - ad_break_fitness: { interruption_risk, break_quality, score }
  Return a complete JSON array over the full timeline.`
});

// Cached to Vercel Blob for instant re-use
await put(`analysis_v3_${videoId}.json`, JSON.stringify(parsed));

// Cosine similarity between ad creative and scene segment vectors
function cosineSimilarity(a: number[], b: number[]): number {
  let dot = 0, nA = 0, nB = 0;
  for (let i = 0; i < a.length; i++) {
    dot += a[i] * b[i]; nA += a[i] ** 2; nB += b[i] ** 2;
  }
  return dot / (Math.sqrt(nA) * Math.sqrt(nB));
}

// Stretch tight cosine range [0.35–0.75] → full [0–1] scale
const rawSim = cosineSimilarity(ad.vector, segment.vector);
const SIM_MIN = 0.35, SIM_MAX = 0.75;
let contextMatch = (rawSim - SIM_MIN) / (SIM_MAX - SIM_MIN);
contextMatch = Math.pow(Math.max(0, Math.min(1, contextMatch)), 1.5);

// Stage 1 — sceneFit: does this ad BELONG in this scene?
const sceneFit =
  suitableMatch  * 0.15 +  // Pegasus suitable_categories hit
  environmentFit * 0.15 +  // Environment × category affinity table
  toneCompat     * 0.10 +  // Emotional tone compatibility
  contextMatch   * 0.60;   // Marengo vector similarity (dominant)

// Stage 2 — adAffinity: pre-computed user → ad category score
const totalScore = adAffinity * sceneFit;

// Diversity: no ad wins twice; category cap = ceil(breaks / 2)
return selectAdsWithDiversity(ranked, plan, config);

-- Generated by buildAdMetadataExportSql()
-- Target: main.default.ad_metadata_premium_spirits
-- marengo_embedding_json: JSON float[] — cast with
--   from_json(marengo_embedding_json, 'array<double>') for Vector Search
CREATE OR REPLACE TABLE `main`.`default`.`ad_metadata_premium_spirits` AS
SELECT * FROM VALUES
  ('grey_goose_30s.mp4', 'Grey Goose — The Art of Enjoyment', 30,
   '["Bar scene","Cocktail","Celebration"]',
   '["Adults","HHI $100K+"]', '["Underage"]',
   '{"highPriority":["Luxury","Premium Spirits"]}',
   '["Children programming","Violence"]', '[]',
   '[0.021,-0.034,0.119,...]', 512, 'twelvelabs_marengo',
   'embedded_marengo_clip_avg')
AS v(creative_id, campaign_name, duration_sec, extracted_visual_contexts,
     target_demographics, negative_demographics, target_audience_affinity,
     negative_campaign_contexts, brand_safety_garm, marengo_embedding_json,
     embedding_dim, embedding_model, vector_sync_status);