Profiles, skills and real-time messaging on one search index

The problem

A professional network needs three things that pull against each other: profiles rich enough to be worth searching, search fast enough to feel instant, and messaging immediate enough that people come back. Built naively, each one degrades the others — richer profiles slow the search, and polling for messages hammers the same database the search runs on.

The approach

Profiles model people and companies separately but share the structures worth searching on — expertise, skills, certificates, languages, location. Search runs against Elasticsearch through a document layer rather than against the relational tables, so adding a field to a profile does not make the search slower. Messaging runs over its own Redis-backed real-time channel instead of polling, and the notification and follow graphs are written asynchronously through Celery.