05
SaaS Marketplace / Real-time Platform

Connectingmallretailerstocustomersathyper-localspeed.

CompanyCityCenterMall
YearCityCenterMall
TypeSaaS Marketplace / Real-time Platform
RoleLead Full Stack Engineer
Scroll
01 — Problem Statement

Mall retailers had no digital presence and orders were pure chaos.

Local mall retailers operated in complete isolation — no unified storefront, no real-time order system, no delivery coordination. Orders came in via WhatsApp calls and were manually logged by store staff. When items sold in-store, digital listings stayed unchanged, causing double-sells. Delivery assignments were handed out manually, creating 15–20 minute matching delays and a 22% order cancellation rate that was destroying merchant trust.

Key Metric latency
22%
Order cancellation rate before platform
02 — Solution Framework

An event-driven WebSocket pipeline with AI-powered catalog management.

Built the full platform from scratch — a persistent WebSocket gateway keeps all three parties (buyer, merchant, delivery partner) in sync with sub-second latency. When an order is placed, a broadcast fires to the merchant in under 200ms. A geo-indexed PostgreSQL engine calculates the nearest available rider automatically, cutting matching time from minutes to seconds. For catalog management, a Gemini-powered PDF parser lets vendors upload raw supplier sheets — it reads, classifies, and creates structured listings across 20+ product categories with zero manual entry. Three delivery modes (Express 60-min, Scheduled, Self-Pickup) run off the same order state machine.

The Hypothesis

By wiring buyers, merchants, and delivery partners into a shared real-time state, the platform eliminated the entire manual coordination layer that was causing a 22% cancellation rate.

Engineering Core Principle

03 — Key Features

Architectural Highlights

01

Persistent WebSocket Gateway

Maintains live socket channels across all platform surfaces with automatic reconnection logic. Order lifecycle events — placed, accepted, dispatched, delivered — broadcast to all relevant parties in under 200ms, replacing the manual WhatsApp coordination entirely.
02

Geo-Proximity Dispatch Engine

A PostgreSQL PostGIS query engine ranks nearby active delivery partners by proximity, current queue depth, and delivery mode. The top match is automatically assigned and notified — no manual dispatcher required. Average rider-assignment time dropped from 15 minutes to under 40 seconds.
03

Gemini AI Catalog Pipeline

Vendors upload raw supplier PDF price lists. The Gemini LLM pipeline parses unstructured data, infers product categories, maps to the platform taxonomy, and creates structured lot listings with pricing, GST codes, and images — completely eliminating the 3-day manual onboarding process.
04 — Results & Impact

Measured outcomes.

80%
Reduction in order lifecycle latency
< 200ms
Merchant notification time
0
Manual catalog entry via AI pipeline
3 Modes
Express, Scheduled, Self-Pickup
Next Case Study

Platz-Halter

06Platz-Halter