Software Engineer

Architected a configurable JSON-driven rule engine for identifying duplicate payment instructions across payment types. The system handled nested conditional validation and generated optimized SQL queries within defined time windows, reducing monthly duplicate payment instructions by 97%.

Developed user-facing Portfolio Accounting Portal features using Java Spring Boot, React 19, TypeScript, and GS UI Toolkit. Implemented OAuth with GS PingFederate, frontend Jest tests, backend JUnit tests, Java 8 to Java 17 migration, centralized secrets integration, and FastAPI trade-data validation endpoints.

Optimized SQL across Db2, Sybase, and Oracle Database, and deployed an on-premises Spring Boot application to AWS ECS Fargate with multi-region support. Worked in an Agile, JIRA-based environment with cross-functional teams.

Added multithreading capability to an internal Apache Kafka library for Java Spring Boot applications, improving message-processing throughput by up to 2.7x.

Built a performance testing suite using Apache JMeter to evaluate Kafka behavior within microservices and validate the reliability of the enhanced implementation.

Built and scaled a Codeforces-style mathematics problem archive and contest platform to more than 30,000 users and 10,000 monthly active users. Led a team of two developers and owned end-to-end backend, frontend, architecture, and infrastructure decisions using Laravel, MVC, and Eloquent ORM.

Reduced database load by approximately 50% through Redis caching and integrated TypeSense full-text search. Delivered real-time updates and notifications with Livewire and WebSockets, while moving intensive PDF generation and rating calculations to queue jobs.

Established multi-environment GitLab CI/CD pipelines, Kubernetes deployments, S3 asset hosting, and production observability with Prometheus, Grafana, and Loki.

Developed a deep learning model using Generative Adversarial Networks to generate front and back normal maps of a human body from a single RGB image.

Improved the PiFuHD approach by addressing blur and occlusion issues through convolution-layer enhancements and optimized data preprocessing, producing a 4% accuracy improvement.

Built an API backend for an online compiler that executed C, C++, and Python code in a sandboxed environment and returned output for supplied input, serving up to 1.8 million requests per day.

Improved fault tolerance and uptime to 99.9% by introducing an NGINX load balancer, separating web and compilation servers, and implementing a task scheduler for parallel workload distribution across servers.