Case Study
A UK furniture reseller was competing for auction stock at a speed manual monitoring could never match, while customer queries piled up. We engineered the full stack behind the operation: distributed scraping, a custom inventory database, a pricing engine, and an AI support agent that captures every lead.
Premium Furniture Reseller
Staff were monitoring multiple auction sites by hand: slow, error-prone, and impossible to scale
High-value furniture pieces were slipping away because listings surfaced too late
Support was buried in repetitive stock and availability questions
No central system to track auction activity or analyse trends
Inventory lived in scattered spreadsheets with no photo management
Pricing was guesswork, with no historical data behind any decision
Engineered a distributed scraping system monitoring 15+ furniture auction sites around the clock
Real-time alert pipeline that flags new listings the moment they match acquisition criteria
AI support agent that resolves product queries, captures every lead, and books viewings
Analytics dashboard surfacing market trends and price movements across the auction landscape
Custom inventory database with photo galleries, condition tracking, and full-text search
Pricing engine that recommends selling prices from historical auction data and seasonal demand
01 / What We Built
Scraping infrastructure covering 15+ auction sites with deduplication, image extraction, and automatic category classification. Nothing gets listed without Sofarama seeing it first.
A multi-step agent that answers stock, pricing, and delivery queries, captures lead details, and books viewing appointments without pulling staff off the floor.
Real-time visibility into auction trends, price movements, and popular styles, feeding predictive pricing recommendations back into the business.
02 / Backend & Data
The data infrastructure running the entire operation: distributed scraping workers, a custom database, and a pricing engine.
PostgreSQL with full-text search, storing every auction find with images, provenance, and condition reports. A purpose-built admin panel handles inventory and photo management.
Node.js scraping workers coordinated through Redis queues, with rate limiting, proxy rotation, and automatic retry logic engineered for reliability across 15+ auction sites.
Price estimation built on historical auction data, weighing style, condition, maker, and seasonal demand to recommend the selling price each piece can actually command.