Blogs

  • Artificial Intelligence

VoomiSupply: From Keyword Search to a Self-Updating HVAC Commerce Platform

Published On: 5 April 2026.By .

VoomiSupply: From Keyword Search to a Self-Updating HVAC Commerce Platform

AI & Data Engineering Case Study
VoomiSupply Cygnus semantic search dashboard showing natural-language HVAC queries matched to intent across 1.5 million SKUs
1.5M+
Product SKUs Processed
80,000+
Enriched HVAC Product Types
Real-Time
Data Enrichment Pipelines
AI
Semantic Search Engine
Cygnus
Admin + Insights Platform
Live
Search & Traffic Analytics
01: The Challenge

A Catalogue That Had Outgrown Its Search

Four gaps were compounding as VoomiSupply scaled: unstructured product data across 1.5M+ SKUs, no automated way to enrich HVAC-specific attributes, a keyword-only search that could not interpret buyer intent, and no unified view of how the platform was actually performing.

Each problem made the others worse. Bad data degraded search results. Poor search results made attribute gaps more painful. And without a central analytics view, the team could not measure any of it.

Keyword-only search returning zero results alongside spreadsheets of missing SEER2 and tonnage attributes across the VoomiSupply catalogue
Before: keyword matching, missing HVAC attributes, and search analytics pulled by hand from separate tools.
01
Unstructured Product Catalogue
1.5M+ listings included incomplete, inconsistent, and weakly classified data. Buyers searching by specification - SEER, BTU, tonnage - were getting unreliable results because the underlying data was not structured to support it.
02
No Attribute Enrichment at Scale
Critical HVAC attributes were often missing or inaccurate across the catalogue. With 1.5M+ SKUs, manual upkeep was impossible - there was no automated pipeline to fill or correct these gaps.
03
Keyword-Only Search
The existing Searchanise plugin matched exact keywords and could not interpret buyer intent, handle technical specification queries, or understand natural-language product descriptions. A buyer searching "2-ton heat pump for small office" got nothing useful.
04
No Centralized Platform Visibility
Search behaviour, product performance, stock distribution, conversion ratios, and traffic quality were tracked across separate tools - or not tracked at all. Stakeholders had no single operational view.
02: The Solution

Five Integrated Capabilities, One Intelligence Platform

Five capabilities were built and integrated: a custom AI semantic search engine replacing Searchanise, automated HVAC data enrichment pipelines, Cygnus Copilot for conversational catalogue admin, Cygnus Insights for real-time analytics, and a purpose-built search layer designed specifically for marketplace-scale complexity.

Rather than extending a plugin not designed for this scale, the entire search and data stack was replaced with a connected system - where enriched product data feeds better search, search behaviour informs analytics, and the admin layer can act on both.

Five integrated Cygnus capabilities: AI semantic search, Cygnus Copilot, real-time enrichment pipelines, Cygnus Insights, and a purpose-built search layer
One connected platform: search, enrichment, conversational admin and analytics built to work together.
1

AI Semantic Search Engine

Buyers can now search in natural language - product type, capacity, use-case, or specification - and get relevant HVAC results instead of keyword mismatches. The engine understands intent rather than just matching text, so a query like "quiet 2-ton mini split for basement" maps correctly to the right product category and attributes. This replaces the keyword-only Searchanise plugin with a purpose-built search layer designed for a 1.5M+ SKU catalogue.

Natural-language product queries Intent-based result mapping HVAC specification understanding Replaces Searchanise plugin
2

Cygnus Copilot - Conversational Admin

The internal team can now manage search behaviour, inspect catalogue coverage, and update search logic through plain-English prompts - without manual tagging, spreadsheet-based updates, or developer involvement for routine operations. Synonym mapping, product intent linking, and catalogue question answering all happen through a single conversational interface. A memory-driven learning layer means the system improves progressively as the team uses it.

Natural-language catalogue queries Search term and synonym mapping Prompt-based admin workflows Memory-driven progressive learning
3

Real-Time Data Enrichment Pipelines

Product records across the full 1.5M+ SKU catalogue are now continuously enriched with structured HVAC technical attributes - SEER, SEER2, BTU, tonnage, zones, and product classifications - without manual upkeep. Pipelines process product and inventory updates automatically as they come in, so the catalogue stays accurate as VoomiSupply's supplier base expands. The enriched data directly improves search relevance and filter accuracy for buyers.

SEER, SEER2, BTU, tonnage enrichment Automated at 1.5M+ SKU scale Continuous ingestion and classification Feeds search relevance directly
4

Cygnus Insights - Real-Time Analytics Dashboard

Stakeholders now have a single place to see catalogue structure, stock distribution, pricing ranges, search trends, add-to-cart behaviour, conversion metrics, and traffic quality - all updating in real time. The dashboard also identifies and controls bot and scraper activity through traffic monitoring and IP blocklisting, giving the team both business intelligence and platform protection in one view.

Product and stock distribution Search trends and conversion metrics Add-to-cart and pricing analytics Bot detection and IP blocklisting
5

Purpose-Built Search Layer Replacing Searchanise

Rather than extending Searchanise - a keyword-matching plugin not designed for marketplace complexity - the entire search layer was replaced with a system built specifically for VoomiSupply's catalogue scale, HVAC product taxonomy, and internal operational needs. The replacement is integrated with the enrichment pipelines and analytics platform, so search quality improves as the data does, and the team can see the impact in real time.

Full Searchanise replacement HVAC taxonomy-aware search Integrated with enrichment pipelines Search quality visible in analytics
03: Technology Stack

What Powered the Platform

Shopify as the commerce foundation, a custom semantic search engine for intent-based discovery, LangGraph and Neo4j for agent orchestration and memory, PostgreSQL with pgvector for semantic retrieval, and real-time enrichment pipelines for automated HVAC attribute quality at scale.
Category Technology Role
CommerceShopifyExisting marketplace foundation for product catalogue and transactions
SearchSemantic Search EngineMaps buyer intent and natural-language queries to relevant HVAC products
Admin AICygnus CopilotPrompt-based interface for search management and catalogue operations
AnalyticsCygnus InsightsReal-time dashboards for catalogue, search, traffic, and conversion visibility
AgentsLangGraph + Neo4jOrchestration and memory relationships for adaptive AI workflows
RetrievalPostgreSQL + pgvectorSemantic retrieval and vector-based product intelligence
PipelinesReal-Time EnrichmentAutomated ingestion, classification, and HVAC attribute enrichment at scale
04: Business Impact

What Changed After Go-Live

VoomiSupply moved from a keyword-driven storefront with incomplete data and no unified analytics to an intent-driven marketplace where search, enrichment, admin, and visibility all work from the same connected platform.
Cygnus Insights dashboard showing attribute coverage, enrichment pipeline queue, traffic quality by source and bot sessions blocked
After: catalogue, search, conversion and traffic quality in one live view, with enrichment running continuously.
Buyers Find Products by Intent, Not Just Keywords
Natural-language queries - by capacity, use-case, or specification - now return relevant HVAC results. Buyers no longer need to know the exact product name or model number to find what they are looking for.
Catalogue Quality Maintained Without Manual Work
HVAC attributes across 1.5M+ SKUs are enriched automatically as new products come in. The team no longer maintains product data by hand - the pipelines do it continuously as the supplier base grows.
Admin Operations Run Through Prompts, Not Spreadsheets
Search term mapping, synonym management, and catalogue queries that previously required manual effort or developer time now happen through plain-English prompts in Cygnus Copilot.
Stakeholders See One Live View of Platform Performance
Search trends, stock distribution, conversion metrics, pricing ranges, and traffic quality are all visible in one real-time dashboard - replacing scattered tools and manual reporting.
Bot and Scraper Traffic Identified and Controlled
Traffic monitoring within Cygnus Insights flags bot and scraper patterns as they occur. The team can act immediately through IP blocklisting rather than discovering the problem after the fact.
Search Quality Improves as the Catalogue Does
Because the search engine, enrichment pipelines, and analytics platform are integrated, better product data produces better search results - and the impact is visible in the analytics dashboard without manual cross-referencing.
05: Frequently Asked Questions

Questions About This Project

What is VoomiSupply?
VoomiSupply is a technology-driven B2B and retail e-commerce marketplace based in Pennsylvania, USA, offering over 1.5 million products across HVAC, plumbing, electrical, and industrial supply categories. Founded in 2019, it serves contractors, technicians, and homeowners through Shopify.
What was the problem with VoomiSupply's product search?
VoomiSupply was using Searchanise, a keyword-matching plugin not designed for a 1.5M+ SKU HVAC marketplace. It could not interpret buyer intent, handle natural-language queries, or support technical specification searches. Combined with incomplete catalogue data, buyers were unable to find products by SEER rating, BTU, tonnage, or use-case description.
How does the AI semantic search engine work?
The custom semantic search engine understands the intent behind a buyer's query rather than matching exact keywords. A search like "quiet 2-ton mini split for basement" is mapped to the relevant HVAC product category and specification attributes, returning accurate results even when the query does not contain the exact product name. It is built on PostgreSQL with pgvector for semantic retrieval.
What is Cygnus Insights?
Cygnus Insights is a real-time analytics dashboard that gives VoomiSupply stakeholders a single view of catalogue structure, stock distribution, pricing ranges, search trends, add-to-cart behaviour, conversions, and traffic quality - including bot detection and IP blocklisting for unwanted scraper activity.
What is Cygnus Copilot?
Cygnus Copilot is a conversational admin layer that lets VoomiSupply's internal team manage search behaviour, map synonyms, inspect catalogue coverage, and answer product intelligence questions through plain-English prompts - without manual tagging or spreadsheet-based workflows. It uses a memory-driven learning architecture that improves outputs progressively over time.
Why was Searchanise replaced rather than extended?
Searchanise is designed for keyword search on standard e-commerce stores. VoomiSupply needed semantic understanding of buyer intent, automated enrichment for HVAC-specific attributes, conversational admin tooling, and real-time analytics - capabilities that cannot be added to a keyword-matching plugin. A purpose-built replacement was the only way to address the full problem.
What AI technologies power the platform?
The platform uses a custom semantic search engine, Cygnus Copilot for conversational admin AI, LangGraph and Neo4j for agent orchestration and memory relationships, PostgreSQL with pgvector for semantic retrieval, and real-time enrichment pipelines for automated HVAC catalogue quality at scale.

Running a Large Catalogue on Keyword-Only Search?

If your e-commerce platform is growing faster than your search and data infrastructure can support, Auriga IT builds custom AI search engines, enrichment pipelines, and analytics platforms designed for catalogue complexity - not generic plugins.

Talk to Auriga IT

Related content

Stay Close to What We’re Building

Get insights on product engineering, AI, and real-world technology decisions shaping modern businesses.

Get regular updates of our latest insights

Services

Products

Resources

© 2026 Auriga. All rights reserved.

16+ Years of Excellence • 1000+ Projects Delivered

Go to Top