Feeds

Personalized feeds, built and deployed in minutes

Build your recommendation algorithm for personalized onchain asset recommendation. Serve personalized feeds via a single API call. 250ms inference.

175M+
ranking calls served
2.5x
engagement lift
250ms
P50 inference
50K
DAUs served
The Problem

Generic feeds are killing your engagement

01

One-size-fits-all feeds don't convert

Every user sees the same content regardless of their interests, trading history, or risk profile. Most of it is irrelevant noise.

02

Building recs in-house takes years

ML pipelines, feature stores, model training, A/B testing infrastructure. Most teams don't have the data science headcount or the onchain training data.

03

Cold start kills new user retention

New users see generic content, get no personalization, and churn before your system learns anything about them. By then it's too late.

Why Feeds

Turn passive browsing into active engagement

Engagement

2.5x engagement lift

Personalized feeds surface the content each user actually cares about. More relevant items = more clicks, more swaps, more time in app.

Retention

A feed worth coming back to

When every session shows fresh, relevant content tailored to their interests, users build a daily habit. Generic explore pages don't create that.

Revenue

Discovery to Fees

Every feed item is a CTA. Users discover tokens, markets, and content they wouldn't have found, and act on it. Direct path to transaction revenue.

How It Works

From idea to production feed in minutes

Build your recommendation algorithm, compose it with other pipelines, and deploy.

Step 1

Build Your Algorithm

Use the Studio SDK to define your search, scoring, and ranking pipeline. Filter candidates, apply ML models, add personalization, all in code.

STUDIO SDK
import { StudioConfig, StudioV1 } from "mbd-studio-sdk";
const mbd = new StudioV1({
config: new StudioConfig({
apiKey: process.env.MBD_API_KEY
})
});
// Search trending prediction markets
const candidates = await mbd.search()
.index("polymarket-items")
.size(50)
.include([
{ filter: "term", field: "active",
value: true },
{ filter: "numeric",
field: "volume_24hr",
operator: ">=", value: 10000 }
])
.execute();
// Score with ML model
const scored = await mbd.scoring()
.model("polymarket-rerank-v1")
.userId("0xab2...e44")
.itemIds(candidates.map(c => c._id))
.execute();
Step 2

Compose & Deploy

Compose multiple algorithms together, preview results, and deploy with a single call. Get a Feed ID that's ready to serve.

DEPLOY
// Compose search + scoring + ranking
// Deploy as a single pipeline
const pipeline = mbd.pipeline()
.search({
index: "polymarket-items",
size: 50,
filters: [
{ filter: "term", field: "active",
value: true }
]
})
.scoring({
model: "polymarket-rerank-v1",
personalized: true
})
.ranking({
strategy: "personalized",
diversity: 0.3
});
// Deploy and get a Feed ID
const { feed_id } = await pipeline.deploy({
name: "trending-markets-feed"
});
// feed_id: "feed_abc123"
Step 3

Serve via API

Pass a wallet address and your Feed ID. Get back ML-ranked, personalized results at 250ms. One SDK call in production.

FEED SDK
import { getClient } from "@embed-ai/sdk";
const client = getClient(
process.env.API_KEY_EMBED!
);
// Serve personalized feed
const feed = await client.feed
.byWalletAddress(
"0xab2...e44",
"feed_abc123",
{ top_k: 25 }
);
Feed Types

One platform, every feed type

Same pipeline, same SDK. Different data indexes and scoring models for each use case.

Trading Feeds

Personalized token discovery based on each wallet's trading history. Surface trending tokens, new listings, and smart money moves, ranked by relevance to each user.

Trending tokens by wallet affinity
New listing alerts
Smart money convergence

Prediction Market Feeds

Recommend markets each user is likely to trade on. Rank by volume, relevance, and past betting patterns across Polymarket and other platforms.

Markets ranked by user interest
Trending categories
High-volume opportunities

Social Feeds

Personalized social content from Farcaster, Lens, and onchain activity. Surface casts, posts, and profiles relevant to each wallet's onchain graph.

Relevant casts and posts
Suggested follows
Onchain social graph
Technology

Purpose-built recommendation models for onchain data

Not a generic ML platform. Models specifically designed for wallet-level behavior and onchain interactions.

Personalization
Sequential, time-aware models trained on each wallet's interaction history. Cold start from just 2 interactions.
Backtesting
Daily validation: mask last 10 actions per user, predict what they do next. No data leakage, no inflated metrics.
Serving
250ms P50 inference via single API call. REST for feeds, webhooks for notifications.
HRNN Architecture
Hierarchical recurrent models capture both short-term intent and long-term preferences from onchain behavior.
130M+ Interactions
Trained on real onchain data: DEX swaps, NFT trades, prediction market bets, social engagement across chains.
Cold Start in 2 Actions
New wallets get useful personalization after just 2 interactions. No weeks of data collection needed.
Multi-Domain
Same architecture works across trades, bets, and social. Cross-domain signals improve recommendations.
Integration

Live in production in under two weeks

1
Get your API key
Sign up at console.mbd.xyz. Free tier available.
2
Build and deploy your feed
Use the Studio SDK or AI feed builder. Get a Feed ID.
3
Call the Feed SDK
One SDK call with wallet address + Feed ID. ML-ranked results in 250ms.
4
Render in your UI
Standard JSON response. Use your own components and design.
Proven at Scale

Battle-tested on real crypto data

Coinbase Base App · Production

Powered the Base App social feed

Our models re-ranked the Base App feed in real time using in-app events + on-chain data. At peak, 50% of transactions originated from personalized feeds, with 2.5x engagement lift.

2.5x
Engagement lift
50%
Tx from feed
100+
Algo recipes
Why This Matters

Years of ML infra, now an API

The largest crypto exchange spent years and a full ML team building user embeddings, a targeting engine, and an ML notification platform. We put the same capability in an API you integrate in two weeks.

"What took a dedicated ML team years to build internally, you can now ship with a single API integration."
FAQ

Common questions

Today we primarily serve prediction market feeds and trading feeds. Rank prediction markets by relevance to each wallet, surface trending tokens personalized to each user's trading history, and recommend assets based on onchain behavior. The platform is composable, so you can combine search, scoring, and ranking to fit your use case.

The largest crypto companies have spent years building internal ML personalization: user embedding models, targeting engines, feature stores, and ranking platforms. That requires a dedicated ML team, streaming infrastructure, and ongoing maintenance. Embed delivers the same capability via API. What takes an internal team years to build, you integrate in two weeks.

Embed's recommendation engine is built on three layers: (1) Wallet embeddings from onchain interactions across protocols. (2) User-item pair features computed at request time for contextual relevance. (3) An HRNN ranking model trained on 130M+ onchain interactions, delivering ranked results at 250ms per API call. Personalization kicks in from just 2 user interactions.

Nothing to get started. Our models are pre-trained on onchain data. For personalization, we use the wallet address your user connects with. No PII, no user accounts. If you want to incorporate in-app events (clicks, views, swaps), you can send those via our ingestion API to improve relevance.

Minimal. Call a REST endpoint or use our TypeScript SDK. Most teams integrate in 1-2 weeks with no changes to existing infrastructure. We plug into your stack, not replace it.

General-purpose tools like AWS Personalize or Algolia Recommend don't understand crypto. They have no concept of wallet behavior, protocol interactions, or onchain trading patterns. Embed is crypto-native: our models are trained on prediction market bets, token swaps, and onchain activity. We understand what a wallet does across the ecosystem, not just within one app.

Yes, that's a core feature. You can chain search filters, multiple scoring models, and ranking strategies into a single pipeline. For example: search trending markets, score by wallet affinity, re-rank by recency, apply diversity rules. Save the whole pipeline as one Feed ID.

Pricing scales with your usage, not your user count. We offer flexible structures depending on your use case and growth stage. Get in touch and we'll find the right fit.

Embed was founded by a former AWS and Chainlink data scientist and is backed by a16z crypto, Polymorphic Capital, Mask Network, Forward Research, WAGMI Ventures, and Social Graph Ventures. The team includes ML engineers who've built personalization and recommendation systems at Amazon, Discovery+, and Bloomberg.

Sign up at console.mbd.xyz to get an API key. Use the Studio SDK to build your first feed pipeline, or try the AI feed builder to generate one from a natural language description. Deploy, get a Feed ID, and start serving with the Feed SDK.

Stop showing every user
the same feed

Build, deploy, and serve personalized feeds with a single API. From idea to production in minutes.

© 2024 ZKAI Labs. All rights reserved.