Event2Vec

Unlocking event sequences with additive embeddings and self-supervised prediction

A running sum of event vectors turns a history of purchases, logs, or clinical codes into something linear models can read directly. That is the deceptively simple idea behind Event2Vec, a small library Antonin Sulc built for self-supervised next-event prediction. The arithmetic is transparent: a churn score becomes a sum of embedded contributions. What starts as root-cause analysis at Berkeley Lab could reshape how we treat sequences as additive, not opaque. For deeper coverage, see how Netflix tackles observability across 38M events per second.