0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilor

dفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy

jmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈

qtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0

x▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بd

◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgj

∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknq

فehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux

ilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇

psvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞

wz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0اف

◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفi

π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmp

cلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtw

hknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆

orux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π

vy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xc

░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افeh

∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilo

بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsv

gjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░

nqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑

ux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1ب

▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلgjmpsvy▒◇∆π1بdفilorux▓◆∇∑0افehknqtwz░◈Ξ∞xcلg

rehan@neural-mesh :~

$ ▊

rehan.osRehan Rao

Search Systems 2025 · 05 11 min

Serving 49M records at sub-100ms.

Query parsing, normalization pipelines, and rank shaping. The full anatomy of a large-scale search backend engineered from scratch.

Rehan Rao

AI & Backend Systems Engineer

architecturesearch.pipeline

49 million rows is not big data. It is the awkward middle — too large to scan, too small to justify a distributed search cluster from day one. The right answer is a boring pipeline with sharp edges.

The shape of the problem

Users type messy strings. The corpus is semi-structured. P95 latency budget is 100ms including network. That budget forces every stage to be measured in single-digit milliseconds.

The pipeline

Every query flows through the same seven stages. The magic is that each stage is pure and independently cacheable.

Index layout

Postgres + trigram GIN for fuzzy prefix.
Meilisearch shard per vertical for typo tolerance and rank features.
Materialized tsvector column, refreshed by change-data-capture.

Rank shaping

BM25 alone ranks nothing users care about. Blend it with recency, popularity, and a learned per-vertical bias. Keep weights in config, not code, so you can A/B-test rank without a deploy.

score = 0.55 * bm25
      + 0.20 * log1p(popularity)
      + 0.15 * recency_decay(created_at)
      + 0.10 * vertical_bias[v]

Serving budget

Parse + normalize: < 3ms.
Shard fan-out: < 30ms P95.
Rank + hydrate: < 40ms P95.
Serialize + wire: < 10ms.

note

Warm the top 1% of queries in a rank cache. On any real traffic distribution this covers 40–60% of hits and drops the tail dramatically.