San Jose, California, United States
You can find more about me on my personal website: https://karanjoisher.github.io/
• Built NVGrid, a foundational library suite for developing advanced cloud storage apps leveraging NVMe, RDMA • Implemented RDMA channel API using IBVerbs to provide high-speed data transfers with minimum latency • Prepared TCP channel for NICs that don’t support RDMA, and Shared Memory channel for services running on the same node to avoid unnecessary buffer copies • Engineered Connection Manager library for establishing channel connections between nodes; used gRPC service framework and Redis for service look-up; and Boost state machine for connection orchestration • Architected Telemetry library to monitor the health and performance metrics of NVGrid components and integrated it with Grafana for visualization • Tech used: C/C++, IBVerbs, Boost, gRPC, Redis, Kubernetes, Docker, Grafana, Bazel, GoogleTest, Jenkins
• Offered guidance on the Discovery Cluster, covering areas like Linux, Slurm, Docker, CUDA, MPI • Managed Concourse workflows to create and roll out Docker and Singularity containers • Worked alongside scientists to design and initiate ML models on multi-GPU setups using PyTorch and TensorFlow
Barclays Mobile Banking APIs ● Designed and developed an ecosystem of APIs for setting up repayment plans for delinquent accounts. ● Gained 5x improvement in API response times by making downstream API calls asynchronous and parallel. ● Implemented a performance logging and monitoring library to track the performance of these APIs. Customer Outcome Testing Application ● Gathered requirements from bank agents to understand how they manually reviewed collection cases. ● Developed a backend API and database components to digitize the manual case review process. ● Closely worked with bank agents to build UI and iterate over the application flow. ● Key Achievements: Reduced manual effort for case reviews by 232 hours/week (5.8 FTEs). AWS Microservices and Cloud Infrastructure Automation ● Developed microservices to create collection cases for delinquent accounts received via Apache Kafka events. ● Created CloudFormation templates to deploy auto-scaling ECS clusters, MSK clusters, RDS, and EC2 machines on the AWS cloud platform. ● Implemented AWS Lambdas for rotating credentials and replicating EBS volumes to backup data. ● Developed features for APIs to automatically fetch the latest credentials from AWS Secret Manager. ● Key Achievements: Single-click infra provisioning; Reduced server costs by demand-based cluster scaling.