Apr 2025 – Present · 1 yr 2 mos · Bengaluru, Karnataka, India · Hybrid

• Building and maintaining high throughput Backend and AI systems actively powering tracking and returns events at a scale of 6k+ req/sec.
• Optimised logging costs bringing down logs from 4TB a week to less than 1 TB per week. This lead to cost reduction of approx. $3000 weekly which translates to $156k annually.
• Lead AI initiative to predict cases of missing service code in parcels tracking data while reaching 90%+ accuracy across carriers. This service code is critical for estimating when the parcel will be delivered. We were able to increase the delivery date accuracy by an average of 10% for 3 of our major carriers.

Jun 2024 – Apr 2025 · 10 mos · Bengaluru, Karnataka, India · Hybrid

• Designed and built delivery-instructions feature for Narvar's parcel tracking system. This involved deprecating existing delivery instruction feature operating as a MVP and replacing it with a more stable system. And integrating it with users preferred delivery instruction stored within our systems.
• Refactored tracking events pipeline, by migrating the event queue from In-house Apache Pulsar to GCP’s PubSub. Benefits include mitigating the overhead of scaling Pulsar to handle spikes in traffic, being able to support proof of delivery images in tracking data and reducing publishing latency from 100+ms to 1ms due to highly optimised PubSub publishing client.
• Led infrastructure readiness for Peak Preparation FY 2024-25, handling 6000 req/sec load.
• Performed load tests across services.
• Estimated min/max replica to handle load.
• Introduced Queue backlog based horizontal pod autoscaling to handle traffic bursts.
• Optimised image upload latency from 150ms to 70ms by switching upload strategy from resumable to normal uploads.
• Fixed memory leaks and pod restart issues across services.
• Played key role in on-call and weekly metrics review to enhance system performance.