On this page
Active used repeatedly Started Jun 2025

Atomize AI: News Intelligence

Turn AI news into searchable summaries, connected topics, and trend signals.

Product overview Demo Go to App Private Repo

Recorded with the Spectra plugin

Start here Learn the retrieval pipeline Follow one sentence through chunking, embeddings, retrieval, and a grounded output. Try the guided lab Architecture Explore the full architecture Run a scale-model pipeline, inspect its gates, and follow an article from ingestion to retrieval. Explore the model
Next.js 16 Groq and OpenAI PostgreSQL pgvector Redis and BullMQ LangSmith OpenTelemetry Prisma
Recorded with Spectra

Atomize turns AI coverage into connected research

Atomize AI ingests AI news, summarizes and classifies articles, and connects the people, companies, products, and ideas they mention. Readers can move from a news feed to hybrid search, topic trends, a knowledge graph, competitive themes, and a daily audio briefing. Each view uses the same article collection instead of asking the reader to rebuild context across separate tools.

Ingestion and analysis run on separate runtimes

Vercel serves the Next.js app and schedules source refreshes. Redis and BullMQ queue article extraction, embeddings, graph enrichment, summaries, and clustering. Railway workers process those longer jobs outside the web request. PostgreSQL stores articles and graph data; pgvector supports similarity search across them.

This split lets the feed publish new items while the heavier analysis continues in the background. It also keeps article processing out of the path a reader uses to browse or search.

Hybrid search combines PostgreSQL text search with vector similarity and model reranking. The graph links extracted entities and relationships across articles. Clustering and burst detection surface topics whose coverage is accelerating; competitive views follow themes across sources.

Graph coverage depends on extraction. The current enrichment path can finish an article without creating an entity mention, so a feed item does not automatically have a graph connection. Admin views expose entity quality for review.

Model routing sets a cost boundary before each call

Atomize routes model work across Groq and OpenAI according to the task. Per-provider spend caps check the budget before a paid call starts. LangSmith traces model behavior, while OpenTelemetry and Sentry help inspect processing and failures. Those controls make model choices and operating cost visible as the news collection grows.