Visana Split: Split PhonePe's 2nd largest monolith into category-specific microservices, routing app and cross-DC ingress traffic to new services with zero-downtime canary rollout (CUG → 10% → 50% → 100%). Zero production bugs, reduced deployment pipeline time by 43% and eliminated cross-category release conflicts (5-6 incidents/month → 0).
ULIP: Leading a team of 3 in E2E backend development of PhonePe's ULIP product targeting 30M+ MAU (₹10Cr+ business). Designed HLD/LLD approved by Architect, driving insurer API integrations across pre-to-post payment lifecycle.
Janus V2: Engineered a configurable workflow orchestration framework with pre-filling, journey resumability and simulations. Serves 10M+ workflows/day at P99 < 250ms, adopted across 2 BUs and 6+ pods, powering 15+ journeys.
Stored Mandates: Executed a cross-cluster MariaDB migration of 600k+ policy mandates, implementing phased dual-write with feature-flagged traffic rollout and row-level checksum validation. Developed mandate lifecycle APIs for Health/Life categories, powering the Collection Engine for at-scale mandate processing.
Product Feature Launches: Boosted Life Insurance payments by 20% and average monthly ticket size from ₹1150 to ₹1320 via Limited Pay feature, onboarding 7 insurers across 5 plans. Built Life status tracking, surfacing real-time policy lifecycle updates across 6 insurers, app, agent console, and ops dashboards.
Open Source: Contributed a session-reuse mechanism to PhonePe's open-source DB sharding bundle, eliminating premature DB session closure for transactions — adopted across 25+ teams at PhonePe.
Eliminated P0 4xx production outages in automated chat system by designing a RabbitMQ backed Golang event processor for rate limited processing. Improved tail-latency P98-P99 of 7k+ TPS legacy MySQL transactions by optimizing indexing, query planning and batching to reduce network RTTs.
Reduced Customer Delight team chat volume by 25% by building a Django based Infra to manage 250k+ premium plan subscriptions and launching in-app features like coach change and plan pause/resume.
Led design and development of a dynamic feature flag in Java and Spring for Payments, eliminating mandatory login and reduced payment completion time by over 60%. Implemented event-driven Kafka APIs handling 20M+ online/offline merchant communication channels.
Decoupled services using Pub/Sub pattern with Python, Redis and Kinesis. The system was able to asynchronously valuate 3M+ Real Estate properties in bulk.
Saved £50k in third-party licensing fees by building an in-house tool rating millions of Central London properties using neighborhood health indicator scores. Improved CI/CD pipelines and AWS Infrastructure, reducing developers on-boarding time from 1 day to 30 mins and accelerating feature release by 60% through 1-click deployments.