BUILDING · Platform
Nexus
Personal intelligence system — trend detection across your own sources.
The problem
By the time a trend is obvious, it's too late to be early. Normal consumption — scrolling feeds, reading threads — surfaces narratives long after the momentum started.
How it works
Detect momentum, not noise
Ingests signals continuously, embeds them locally, and clusters them into narratives — then scores each on velocity, acceleration, novelty, and cross-source confirmation.
A momentum score, every 20 minutes
A pipeline cycle runs on a schedule and tags each narrative emerging / accelerating / peaking / cooling, ranked by a composite score.
Embeddings-first, LLM-sparing
Built on pgvector + quantitative momentum math instead of burning tokens — fast, cheap, and yours. Ships with a dashboard of ranked narratives and the full breakdown.
● In private development — FastAPI + pgvector, local embeddings (MiniLM), momentum scoring.