Documentation

How It Works

Bloom.ai turns raw global data into personalized intelligence. A 3-layer AI pipeline ingests, correlates, and ranks events — while an engagement system adapts to each user in real time.

From raw data to actionable intelligence

Every piece of data passes through a pipeline that classifies, correlates, and scores it before it reaches any user.

Step 1

Ingest

News, markets, crypto, and geopolitical data collected from global sources

Step 2

Classify

AI enriches each item with categories, severity, entities, and embeddings

Step 3

Correlate

Related items are clustered into unified events using vector similarity

Step 4

Deliver

Personalized feeds, trends, predictions, and AI chat powered by scored events

Data sources

News / RSS
Stock Markets
Crypto
Geopolitical

Deep dives

Explore the two core systems that power the platform.

Data Pipeline

How raw data is ingested from four sources, classified with AI, and correlated into unified events using embedding similarity.

  • -4 data sources, one unified schema
  • -AI-powered classification and entity extraction
  • -768-dim embeddings via Google Gemini
  • -Cosine similarity event correlation
Read more

User Engagement

How interactions are tracked, interest vectors are computed, and personalized feeds are ranked for each user in real time.

  • -Implicit signal tracking with weighted scores
  • -Time-decayed interest vectors
  • -Hybrid actionability scoring formula
  • -Smart notifications and adaptive ranking
Read more

Built with

The core technologies behind the platform.

Next.js

Full-stack React framework handling both the dashboard UI and all API routes.

PostgreSQL + pgvector

Relational storage with vector similarity search for 768-dimensional embeddings.

Multi-provider AI

Provider-agnostic LLM client supporting Anthropic, OpenAI, xAI, DeepSeek, and Google.

Google Gemini Embeddings

text-embedding-004 generates 768-dim vectors for events, queries, and user profiles.