Quickstart: Studio SDK

This example shows how to use the Embed Studio SDK to build a personalized Polymarket prediction market feed. It demonstrates the full pipeline: search → features → scoring → ranking.

Install

npm install

This installs:

• mbd-studio-sdk — Embed Studio backend client for search, features, scoring, and ranking
• dotenv — Loads environment variables from .env

Configure

Create a .env file in this directory with your credentials:

API_KEY=your-mbd-api-key
POLYMARKET_WALLET=0xYourPolymarketWalletAddress

Both are required. The script will throw an error if either is missing.

Run

npm start

Or directly:

node polymarket.js

The script prints the top 10 recommended Polymarket items with their text and ranking score.

Algorithm Step by Step

The demo runs a 5-stage pipeline to produce a personalized, diversified feed of Polymarket markets.

1. Initialize SDK and set user context

const config = new StudioConfig({ apiKey: apiKey });
const mbd = new StudioV1({ config });
mbd.forUser("polymarket-wallets", polymarketWallet);

• Create StudioConfig with your API key.
• Create StudioV1 instance.
• Call mbd.forUser("polymarket-wallets", polymarketWallet) to set the target user for personalization. This drives user-specific boosts and affinity scores later.

2. Search (candidate retrieval)

const candidates = await mbd.search()
  .index("polymarket-items")
  .includeVectors(true)
  .include()
  .numeric("volume_1wk", ">=", 10000)
  .exclude()
  .term("closed", true)
  .term("price_under05_or_over95", true)
  .boost()
  .groupBoost("polymarket-wallets", "ai_labels_med", polymarketWallet, "label", 1, 5, 10)
  .groupBoost("polymarket-wallets", "tags", polymarketWallet, "tag", 1, 5, 10)
  .execute()
mbd.addCandidates(candidates);

• Index: polymarket-items
• Include vectors: Yes (for semantic search and diversity).
• Include filter: volume_1wk >= 10000 — only markets with at least $10k weekly volume.
• Exclude filters:
  • closed = true — skip closed markets
  • price_under05_or_over95 = true — skip markets with extreme prices (too settled)
• Boosts (personalization):
  • groupBoost on ai_labels_med — boost items whose AI labels match the user's labels (weights 1, 5, 10).
  • groupBoost on tags — boost items whose tags match the user's tags (weights 1, 5, 10).

Result: a candidate set of Polymarket items that pass filters and are boosted by user affinity.

3. Features (enrichment)

const features = await mbd.features("v1")
  .execute()
mbd.addFeatures(features);

• Call mbd.features("v1").execute() to fetch signals and metadata for each candidate.
• mbd.addFeatures(features) attaches enriched features to candidates (e.g. topic_score, user_affinity_score).

These features are used in the ranking stage.

4. Scoring (reranking model)

const scores = await mbd.scoring()
  .model("/scoring/ranking_model/polymarket-rerank-v1")
  .execute()
mbd.addScores(scores, "ranking_model_polymarket_rerank_v1");

• Run the Polymarket rerank model: /scoring/ranking_model/polymarket-rerank-v1.
• mbd.addScores(scores, "ranking_model_polymarket_rerank_v1") attaches the model scores to each candidate.

5. Ranking (final ordering)

const ranking = await mbd.ranking()
  .sortingMethod('mix')
  .mix("topic_score", 'desc', 40)
  .mix("user_affinity_score", 'desc', 40)
  .mix("rerank_polymkt1", 'desc', 20)
  .diversity('semantic')
  .lambda(0.5)
  .horizon(20)
  .limitByField()
  .every(10)
  .limit("cluster_1", 1)
  .execute()
mbd.addRanking(ranking);

• Sorting: mix — combine multiple signals with weights.
• Mix weights:
  • topic_score (desc) — 40%
  • user_affinity_score (desc) — 40%
  • rerank_polymkt1 (desc) — 20%
• Diversity: semantic with lambda = 0.5 — balance relevance and diversity.
• Horizon: 20 — consider top 20 items for diversity.
• Limits:
  • limitByField().every(10).limit("cluster_1", 1) — at most 1 item per cluster in every 10 results.

Result: a final ranked feed stored in the SDK context.

6. Output

const feed = mbd.getFeed();
for (const item of feed.slice(0, 10)) {
  console.log(item._id, item._source.question, item._ranking_score);
}

• mbd.getFeed() returns the ranked candidates.
• The script prints the top 10: _id, question, and _ranking_score.

Pipeline Overview

User context (wallet)
        ↓
   [Search] → candidates (filtered + boosted)
        ↓
  [Features] → topic_score, user_affinity_score, etc.
        ↓
  [Scoring] → rerank model scores
        ↓
  [Ranking] → mix + diversity + limits
        ↓
   Final feed (top 10 printed)