Engineering 10-Minute Hyperlocal Delivery at Scale in Indian Metros

How we architected geospatial dispatch engines and dark store inventory synchronization to deliver 2.5 lakh daily grocery orders across Bengaluru, Delhi NCR, and Mumbai.

Quick-commerce in Indian tier-1 metros like Bengaluru, Mumbai, and Delhi NCR operates under brutal physical and network constraints. When a user adds a packet of milk or groceries to their cart on apps like InstaZip, Zepto, or Blinkit, the entire dispatch orchestration must execute in under 50 milliseconds.

The Dark Store Synchronization Bottleneck

The biggest engineering hurdle in 10-minute delivery is not the delivery rider—it is inventory race conditions. During evening peak hours between 7 PM and 9 PM, thousands of concurrent users are browsing inventory from the same hyperlocal dark store. If two users check out the last remaining item simultaneously, traditional relational database locking creates cascading API timeouts.

At Codemind Studio, we solved this by implementing an event-driven inventory engine using Apache Kafka and in-memory Redis data structures. Stock availability is reserved atomically at the cart level using Redis RedisJSON scripts, while background Go microservices asynchronously reconcile the persistent PostgreSQL ledger.

Geospatial Matchmaking with Redis Spatial

To ensure delivery within 10 minutes, packing at the dark store must take less than 90 seconds, leaving 8 minutes for transit. We utilize Redis Spatial indexing combined with live Google Maps traffic telematics to dynamically assign orders to delivery partners who are within a 500-meter radius of the dark store loading bay before the order is even finished packing. This predictive allocation eliminated rider congestion outside dark stores and boosted on-time delivery rates to 99.4%.