Skip to content
All projects
ShippedPlatform · Search· 2022

Global Search Platform

RediSearch implementation.

<50ms
typical query
Typo-tolerant
relevance
Decoupled
from primary DB

The problem

Search was slow and inconsistent, bolted onto the primary database. Every query competed with transactional traffic, and results felt stale.

Context

I designed a dedicated search platform so search became fast, relevant, and independent of the operational store — and reusable across teams.

Architecture

A clean indexing pipeline feeds RediSearch, which serves faceted, typo-tolerant queries in well under 50ms, fully decoupled from the transactional database.

  • Indexing pipeline that keeps the search index fresh.
  • RediSearch for typo-tolerance, faceting, and speed.
  • Decoupled from the operational store to protect both.
  • A reusable platform other teams index into.

Technical challenges

Index freshness vs. write load

Keeping the index consistent with a fast-moving source of truth without overloading either system required a carefully tuned pipeline.

Engineering decisions

RediSearch over bolting onto the DB

A purpose-built search engine gave relevance features and latency the primary database could never match.

Technologies

RediSearchRedisNode.jsTypeScriptGoogle Cloud

Results

Sub-50ms, typo-tolerant search at scale — and a platform other teams could build on. Part of the work recognised as Best Software Engineer of the Year.

Lessons learned

  • Use the right tool; databases make mediocre search engines.
  • Decoupling protects both systems and unlocks features.

What I’d improve today

  • Semantic / vector search layered on top of keyword relevance.