Identity Orbit
An identity-resolution prototype, based on entity resolution work I've done at the day job. Mixes deterministic, probabilistic, and behavioral linking with a proximity score that decays over time.
What it is
A side project on large-scale identity resolution and behavioral linking, shaped by problems I've actually run into at work. It uses deterministic, probabilistic, and behavioral signals together. The proximity score is what holds it all up.
Why
Off-the-shelf entity-resolution systems generate too many duplicates and merge on hard rules. They ignore behavioral history. They treat identity as a one-shot match decision. No decay. No strengthening. No memory of how a profile drifted over time.
Identity Orbit treats identity as a living graph instead. Trust updates as evidence does. Behavioral signals are first-class, sitting next to the deterministic and probabilistic ones.
Status
Designed for billion-record weeks. Merges are explainable. Low-confidence ones go to a review queue instead of quietly corrupting the warehouse.
No public link yet. If you want to talk about how it works, drop me a line at vkavali10@gmail.com.