AWS Aurora Security

Built a platform for DBAs to create test databases using Kubernetes, Rook, and Ceph. Reduced startup time to <1 minutes by leveraging snapshots and restores.

Optimized a database instance controller by migrating the CloudWatch metrics publisher from Java to Rust which, according reduced resource utilization by 10%

I achieved 99.9% data resilience during Kafka failure by sending messages using a Fluentd aggregator service which caches messages and retries if the destination is unresponsive. I designed the service architecture and configuration, creating a design document that was reviewed by executives and obtained security approval. The system was then deployed the service to K8s and I monitored it with Prometheus and Grafana to ensure guarantees were met.

As Wish moved towards a microservices architecture, I built scripts to analyze and convert between Python ODM models, JSON schemas, and mongoDB collections used by developers across multiple teams to assist transition.In addition, I increased transparency into AWS SQS proxy service by adding error metrics and also added an idempotency feature

I built and optimized customer behaviour model using PyTorch, saving an estimated $1m in fraud dollars per year. My work included adding side features to Matrix Factorization recommendation model and adding a Linear Regression meta-model to solve the cold start problem. I interacted with millions of rows of data stored in Snowflake database using Dask and pandas dataframes. To speedup training and optimizing the model, I also implemented Hogwild and distributed GPU processing methods. The model was deployed to AWS EC2 instances on the in-house cloud compute server.

I streamlined data collection by implementing single page application with Vue.js, saving >50% of work/time required to write multinational policies. The application was backed by a GraphQL API, written with TypeScript and Node.js, which stored and queried data in a PostgreSQL database. I also automated testing and deployment to Kubernetes cluster by containerizing apps with Docker and using GitLab’s Auto DevOps CI/CD.