SPSanmarg Paranjpeinsanmarg.hashnode.dev·Jun 14 · 8 min readAI Model Versioning and Rollbacks in Production: Building Recoverable AI Systems at ScaleAlright, let me tell you something that still keeps me up. We've gotten really good at shipping software. Like, really good. Git, CI/CD, automated tests, canary deploys, blue-green releases, kubectl r00
SPSanmarg Paranjpeinsanmarg.hashnode.dev·Jun 2 · 6 min readDetect Cryptominers on Kubernetes with Custom Falco RulesThink of a scenario a cryptominer landed in your staging cluster. It ran for sometime before anyone noticed, not because monitoring was missing, but because the alerts were too generic to trigger a re10
SPSanmarg Paranjpeinsanmarg.hashnode.dev·May 31 · 9 min readHow I Built a Zero-Trust LLM Security ProxyA few weeks ago, I found myself staring at a problem that kept getting worse. Every team in the org was spinning up their own LLM integrations — some through OpenAI, a few through Ollama, one team eve10
SPSanmarg Paranjpeinsanmarg.hashnode.dev·Feb 25 · 5 min readDeep Dive on Distroless + Multi-Stage Build PatternsMost people switch to distroless because someone on Twitter said it’s “more secure.” That’s not engineering. So I did what engineers should do I built the same Go service four different ways, scanned 10
SPSanmarg Paranjpeinsanmarg.hashnode.dev·Feb 23 · 6 min readYour Pods Are Healthy. Your Service Is Correct. Your Ingress Is Fine. So Why Are You Still Getting 5xx?You deploy your application. Pods are Running Readiness probes are Passing Service selectors are correct Endpoints object is populated Ingress is configured properly TLS is valid Yet your ingr00
