The official website of VarenyaZ
Logo

Smart Search — Semantic Discovery

Turn misspelled queries, vague descriptions, and natural‑language questions into instant “that’s exactly what I wanted” moments. Our vector‑powered search engine reads intent—not just keywords—slashing zero‑result pages, doubling conversion for searchers, and driving a 60 % jump in search‑attributed revenue.

Industry

E‑commerce & SaaS Search · Knowledge Management · Gen‑AI

Service

Vector‑Search Architecture · Gen‑AI Retrieval + Rerank · Multilingual NLP · Search Merchandising Suite

Team Setup

1 Product Lead · 2 Search Architects · 4 ML Engineers · 3 Frontend Devs · 2 Data Scientists · 2 QA Specialists · 3 DevOps

Timeline

9 Months

Story

Goal

Build an intent‑aware, sub‑second search experience that would:

  • Cut zero‑result rate by ≥ 20 % (Dropbox saw 17 % with semantic search).
  • Boost searcher conversion at least 2× over site average.
  • Auto‑surface answers, categories, and bundles—no manual synonym lists.
  • Deliver results in < 150 ms P95 across 10 language locales.

Challenge

Keyword legacy and massive catalogs hindered accurate results:

  • Keyword legacy — Boolean match missed slang & compound queries.
  • Catalog sprawl — 1.4 M SKUs, 12 languages, daily delta > 50 K items.
  • Zero‑result rage — 9.6 % sessions ended empty; bounce soared to 42 %.
  • Cold‑start vectors — new items lacked embedding context.
  • Facet explosion — 7 K attributes needed dynamic ranking.
  • Latency budget — success required < 150 ms including rerank.

Our Approach

01

Discover

Shadowed merchandisers & ran 80 K anonymised query audits; found 53 % were natural‑language or misspelled.

02

Design

Prototyped hybrid BM25 + vector recall; user‑tested semantic facets; tuned embeddings on domain vocabulary.

03

Deploy

Deployed on GCP GKE; real‑time embedding via GPU autoscale; blue/green index swaps nightly.

Challange

The Mountain to Climb

Staging a global vector index that’s always fresh and always fast:

01

1.4 M‑item vector index

Replicated across three regions for global performance.

02

Streaming ingest < 45 s

From PIM change to searchable in near real‑time.

03

Multilingual query → English pivot

M2M‑100 handles on‑the‑fly translation to EN embeddings.

04

Compliance: GDPR

“Right to be forgotten” must remove data within 60 s.

05

Per‑tenant synonyms & rules

Multi‑client architecture overlays brand preferences.

06

Facet relevance learning

No large labeled data sets; incremental feedback approach.

Additional Hurdles

Scale ingestion from 50 K daily SKUs; no re‑index meltdown.

Nightly blue/green index swaps; sub‑second failover needed.

Ensure data security, multi‑tenant separation, & compliance logging.

These complexities pushed us to engineer a resilient, multilingual search stack that meets every brand’s operational and compliance demands.

Key Modules Engineered

Covering recall, rerank, personalization, and beyond—every module orchestrates frictionless discovery.

Vector Recall Engine

Approx‑NN (HNSW) + BM25 hybrid; recall +38 %.

Semantic Spell & Stemming

Understands “nikes for trail runnin’.”

Intent Classifier

Detects FAQ vs SKU lookup → dynamic widgets.

Rerank (LLM)

GPT‑4o reranks top 50; MRR +12 %.

Dynamic Facets

Learns facet order per query; CTR +18 %.

Zero‑Result Rescue

Embedding‑expansion fallback; ZRR −31 %.

Multilingual Pipeline

10 locales, pivot → EN embeddings.

Auto‑Synonym Miner

Mines click‑logs for new synonyms nightly.

Query Analytics Hub

Heat‑map & long‑tail gold mine for merch teams.

Personalised Boosting

Vector merge with user profile; CR +9 %.

Voice & Vision Search

ASR for voice, CLIP embeddings for images.

Edge‑Cache Accelerator

CDN‑side candidate cache; 40 % queries served in 20 ms.

User Research Insights

Shoppers who use on‑site search convert up to 50 % higher than average. Home

67 % abandon after two failed searches; semantic rescue retained 42 % of them.

“I love that it understands eco‑friendly running shoes without filters.”

— beta tester

Technology Stack

Frontend
React 18TypeScriptWebGL facets
Search Core
OpenSearch + Vespa vector nodesFaiss HNSW
LLM Rerank
OpenAI GPT‑4o with RAGfallback MiniLM‑L6 on‑prem
Data
BigQuerydbtAirflowKafka CDC ingest
Middleware
Go gatewaygRPCRedis Streams
Infra
GKE AutopilotGPU spot poolTerraformArgoCD
Security
OAuth 2fine‑grained ACLGDPR purge API

A/B - Test Wins

Semantic vs keyword search
Lift: +21 % revenue/querySample: 1.1 M sessions100 %
Zero‑result rescue
Lift: −31 % exitsSample: 300 K queries100 %
Dynamic facets vs static
Lift: +18 % facet‑click CTRSample: 250 K sessions80 % rollout

ROI / Business Impact

Payback in 6 months
Incremental GMV covered build + infra.
Search‑attributed revenue +60 % YoY
Improved recall and semantic rank fuel growth.
Return rate −9 %
Better item fit from intent matching.

Outcome

From guesswork to guided discovery—one search engine now powers global multi‑brand revenue lifts.

Revenue & Growth

  • Searcher conversion 2.1 × site average, adding $84 M GMV.
  • Long‑tail sales +33 % from semantic expansion.

Shopper Experience

  • Zero‑result rate −31 %, bounce −22 %.
  • Average time‑to‑first‑click 1.2 s → 0.6 s.

Operational Efficiency

  • Merch team freed 10 hrs/week—no manual synonym work.
  • Catalog ingestion pipeline now 45 s vs 15 min baseline.

Brand Impact

  • Featured by Algolia as “Top Semantic Search Launch 2025.”
  • Net Promoter Score on search widget +16 pts.

Feature Highlights

1

Vector Recall Engine

intent first
Relevance
2

LLM Rerank

quality second
Precision
3

Zero‑Result Rescue

save the bounce
Retention
4

Dynamic Facets

click‑magnet filters
Engagement
5

Auto‑Synonym Miner

no manual lists
Automation
6

Multilingual Pivot

10‑locale reach
Globalization
7

Personalised Boost

profile relevance
Personalization
8

Voice & Vision Search

hands‑free find
Accessibility
9

Edge‑Cache Accelerator

20 ms hits
Speed
10

Query Analytics

insight gold mine
Optimization
11

Intent Classifier

FAQ vs SKU
Context
12

Image‑first Embeddings

snap‑to‑shop
Innovation
13

Real‑time Inventory Tie‑in

only live stock
Accuracy
14

Merch Rules GUI

boost & bury in clicks
Control
15

Security + GDPR Purge

sleep easy
Compliance

Want to turn every query into a sale?

Book a discovery call—we’ll plug in your catalog, stand up semantic search, and prove the uplift in a 30‑day pilot.

We are committed to a secure and safe web

At VarenyaZ, we use cookies to enhance your browsing experience on our website. You can choose to accept or reject cookies.