Case Study: Swiggy cuts prediction latency in half with Akka

Swiggy cuts P99 prediction latency from 144ms to 71ms with Akka

The customer, Swiggy, a leading food-delivery platform, faced a challenge with its AI-driven order-assignment flow. Its system had to score thousands of order-and-delivery executive pairings across an entire city and return a decision within a strict one-minute SLA. High P99 prediction latency of 144ms and duplicate computational work were threatening this SLA and degrading dispatch quality during peak demand. To solve this, they used the Akka Optimize solution from vendor Akka.

By building its Data Science Platform on Akka, Swiggy implemented several optimization strategies including micro-batching and feature reuse. The solution from Akka dramatically reduced P99 prediction latency from 144ms to 71ms, a 2x improvement, while sustaining over 5,000 predictions per second. This halved the compute cost per prediction and ensured city-scale order assignments were completed inside the critical one-minute window, improving overall delivery operations.

Swiggy