Three-day vendor onboarding and over-selling during flash sales were blocking growth.
Wholesale merchants manage inventory in bulk 'lots' — each with quantity, size breakdowns, pricing tiers, images, and GST/HSN codes. Listing a single supplier catalog manually required a team member spending 3 full days entering data. With vendors unable to onboard fast enough, the platform couldn't scale. During high-demand flash sales, multiple buyers would simultaneously select the same lot — the system had no concurrency protection, resulting in over-sells, cancellations, and merchant trust erosion. There was no automation anywhere in the catalog or inventory pipeline.
Gemini AI invoice parsing with Redis-based distributed lot reservation locks.
Engineered a 4-surface marketplace (Customer App, Vendor App, Admin Panel, Marketing Site) with an AI-powered catalog core. Vendors now upload raw Excel or PDF price sheets — a Gemini AI pipeline reads the unstructured data, infers product categories, maps sizes and quantities to the platform schema, and creates fully structured lot listings with pricing, GST/HSN codes, and categorization automatically. What took 3 days now completes in seconds. For concurrency, Redis-based distributed locks are applied at the lot-reservation layer during checkout — when a buyer claims a lot, a lock is acquired for a configurable TTL. No concurrent buyer can claim the same units during that window, even under peak traffic, guaranteeing inventory integrity.
“AI invoice parsing compressed 3-day manual onboarding to seconds, while Redis lot-reservation locks eliminated over-selling entirely — enabling flash sales at scale with zero cancellation conflicts.”
Engineering Core Principle