Every GPU, Accounted For

GPU allocation across GCP, Azure and Oracle Cloud is a scheduling problem where being wrong is expensive and being slow is worse. The internal platform tracking it had to read and write at conversational speed.

I led a team of six engineers building it, and owned the frontend architecture.

The frontend is Next.js 14 — React Server Components and edge functions — with Redux for client state, Tailwind for the system, and Headless UI as the accessible primitive layer.

Low-latency reads and writes come from optimistic updates and layered caching, so a real-time workflow feels immediate instead of merely fast.

The data-heavy surfaces are the platform: advanced tables with complex multi-dimensional filtering, and interactive charts built to WCAG standards that were adopted as the baseline across the product.

I integrated a FastAPI and Azure OpenAI conversational assistant into the frontend so allocation and delivery-timeline questions could be asked in plain language.

I owned CI/CD and the PostHog product-analytics infrastructure, and worked daily with design in Figma and with remote stakeholders through agile ceremonies.

The accessibility standards set on these charts and tables became the platform-wide default.