Jay Shah

At Gridwise I work on low-level GPU compute, building and tuning WebGPU kernels for data-parallel workloads. The role is deeply technical — close to the metal, focused on squeezing real throughput out of GPU hardware.

• Built and optimized GPU kernels for scan, reduction, radix sort, and histogram using WebGPU, following CUDA-style execution patterns with thread groups, shared memory, and barriers.
• Profiled kernels with WebGPU tooling, identifying bandwidth bottlenecks and atomic contention; tuned memory layouts to achieve sustained throughput of 80 GB/s — a 40 GB/s improvement over baseline.
• Unified multiple histogram implementations into a single, easy-to-use GPU API, simplifying integration for other developers without sacrificing performance.
• Performed end-to-end performance testing across workloads, analyzing occupancy, synchronization costs, and memory efficiency, iterating on kernels to maximize throughput.

My first industry role, working on embedded firmware and IoT systems for real-world environments. I wrote C++ on STM32 microcontrollers, interfacing with multiple sensor types and building reliable data pipelines under strict hardware constraints.

• Developed a multi-sensor IoT monitoring system on STM32 using C++, integrating temperature/humidity, air quality, and IMU sensors for real-time environmental data acquisition and analysis.
• Designed and optimized a real-time data logging and alerting pipeline on STM32, triggering threshold-based notifications for anomalies such as gas leaks, temperature spikes, and abnormal vibration events.