Noor Siddique

During nationwide promotional campaigns, millions of users flooded the PayPay app simultaneously. These massive traffic spikes created extreme read/write contention on our central databases, driving primary CPU utilization above 95% and threatening platform stability at peak moments.

To safeguard system availability, I designed and implemented dynamic read/write split database routing mechanisms directly at the microservice layer. By analyzing query patterns across Java and Spring Boot applications, we seamlessly offloaded heavy read traffic, profile lookups, and token validations to read replicas, reserving primary database power strictly for critical state writes.

The outcome: This architectural shift completely eliminated CPU spike bottlenecks during high-stakes marketing events, guaranteeing 99.99% system availability and a smooth, low-latency experience for tens of millions of daily users.

Explosive user growth on our core authentication platform brought an unseen bottleneck to light: rapid memory exhaustion in our Redis session caching clusters. High-volume user sessions were relying on native Java JDK serialization, generating heavy binary payloads that strained memory limits and threatened cache response times under peak loads.

Rather than simply throwing expensive infrastructure at the problem, I led a deep technical overhaul of our caching strategy. We re-engineered the session token serialization pipeline by migrating from JDK serialization to Protocol Buffers (Protobuf). I restructured high-concurrency session data models within our Java and Spring Boot microservices to optimize payload encoding and binary efficiency.

The outcome: The migration drastically compressed session payload sizes and slashed Redis RAM consumption across production clusters, preserving sub-millisecond cache latency and saving substantial cloud infrastructure overhead.

As PayPay scaled past 70 million users, our identity infrastructure reached a critical crossroads. Core authentication was tightly coupled, creating architectural friction as we expanded into group ecosystem partners like PayPay Card and PayPay Bank. As engineering manager and lead architect, I set out to transform our core user module into a unified, group-agnostic multi-tenant platform.

Navigating this meant aligning technical vision across 30+ engineering groups operating over 200 microservices without breaking existing dependencies. I spearheaded the architecture of multi-tenant domain isolation boundaries and streamlined low-latency OAuth2/OIDC verification flows in Java and Spring Boot.

The outcome: We successfully decoupled core identity systems, enabling seamless cross-app partner integration while serving 70+ million registered users with zero system downtime.

Driven by a passion to solve operational friction for retail businesses, I set out to build DokanMate (dokanmate.com)—a real-time inventory and stock management platform designed for speed, reliability, and multi-user data consistency.

Building a production-grade platform end-to-end requires balancing rapid feature iteration with robust back-end architecture. I leveraged cutting-edge AI workflows to accelerate code generation, automate boilerplate development, and rapidly prototype complex UI components.

Acting as the primary lead architect and product manager, I directed AI toolchains to generate high-performance data pipeline templates, refine state-synchronization protocols, and speed up full-stack integration. I focused my core engineering effort on optimizing database indexing, eliminating race conditions during concurrent stock adjustments, and enforcing clean architectural boundaries.

DokanMate evolved into a live, production-grade web application in record time. It demonstrates how modern AI-assisted engineering, paired with senior architectural leadership, can rapidly deliver robust systems that eliminate inventory drift for store owners.

This is my own product. I developed the first version as an offline desktop application in 2011. Six customers are still using the desktop version, and I created the second version as an online browser-based application in 2020. Currently, the system is being used by three customers.

I developed an online prescription system for doctors in Bangladesh, addressing challenges with handwritten prescriptions that were often difficult to read and prone to errors. The system also preserved patients’ medication history, improving accuracy and continuity of care.