Infrastructure, platforms, and systems — proven by data, not assumptions.
I’m Hirak. With an MSc in Computer Science, I engineer the foundations of Kubernetes, cloud infrastructure, and AI systems.
I build production-grade platforms, break them on purpose, measure exactly how they behave, and document the mechanics of failure and recovery.
Selected work
Projects built to explore the engineering behind reliable, scalable systems.
A production-style platform for deploying and operating AI applications on Kubernetes, with infrastructure as code, GitOps delivery, observability, autoscaling, and reliability controls.
An incident-analysis system that collects Kubernetes logs, extracts relevant failure evidence, and generates structured incident reports containing likely root causes, affected components, and remediation suggestions.
An OpenAI-compatible inference gateway exploring request routing, rate limiting, caching, autoscaling, observability, and the operational characteristics of serving AI workloads.
From the lab
Engineering notes, experiments, benchmarks, failures, and things I learned while building systems.
Case study · Platform Engineering
Designing a static-first publishing platform for technical writing — and treating information architecture, typography, accessibility and performance as one system.
ReadExperiment · Kubernetes
I load-tested two Kubernetes autoscaling approaches and measured how quickly each reacted as traffic moved from idle to sustained load.
ReadBenchmark · Docker
How much does the underlying CPU actually matter when building container images?
ReadNote · Kubernetes
Ingress is much easier to understand when you stop treating it as a Kubernetes-specific concept.
ReadCurrently
Scheduling, networking, autoscaling, failure recovery and platform design.
Developer platforms, GitOps, CI/CD and infrastructure automation.
Metrics, logs, traces, SLOs and incident response.
Inference serving, request routing, caching, performance and cost.
Status
HIRAK is being rebuilt. This site is temporarily locked while the new version is finished.
