Writing

Long-form, by the people building.

Migration patterns, production LLM behaviour, zero-trust inside a cluster. Written by engineers, at the length the subject actually needs, with the parts that went wrong left in.

AI EngineeringFeatured

Generative Engine Optimization: Getting Cited by AI Search

How ChatGPT, Perplexity, and Google AI Overviews choose what to cite, and the markup, structure, and crawl decisions that decide whether your pages are in the answer.

Aaradhy Raghav DuveyAug 05, 202610 min read

Everything we have published

Newest first. No listicles.

AI Engineering

RAG That Survives Production: Chunking, Reranking, and Eval

Retrieval quality, not model choice, is what decides whether a RAG system is useful. The chunking, reranking, and evaluation decisions that separate a demo from a system people trust.

Jul 29, 202612 min read

Architecture

Postgres at 10 TB: What Breaks First, and in What Order

Connection exhaustion, autovacuum falling behind, and index bloat arrive in a predictable sequence as a Postgres database grows. Here is the order, and what to do at each step.

Jul 08, 202611 min read

Cloud Infrastructure

Cutting a Kubernetes Bill by 40% Without Touching Reliability

Most Kubernetes overspend is idle capacity you requested and never used. How to find it with the data you already have, and the four changes that recover the most.

Jun 17, 20269 min read

Cloud Infrastructure

Enforcing Zero-Trust in Multi-Tenant Kubernetes

A deep dive into NetworkPolicies, Istio sidecars, and enforcing strict mutual TLS between internal container workloads.

May 21, 202611 min read

Architecture

Multi-Tenant SaaS: Shared Schema, Schema-per-Tenant, or Database-per-Tenant

The tenancy model you pick on day one decides your migration story, your compliance answers, and your per-customer cost for the life of the product. A decision framework with the trade-offs stated plainly.

Apr 30, 202610 min read

AI Engineering

LLM Orchestration: Managing Latency in Production

Strategies for managing rate limits, semantic caching, and token streaming when deploying Large Language Models at scale.

Apr 09, 20266 min read

Web Performance

INP Is the Metric That Finally Catches Slow React

Interaction to Next Paint measures every interaction on the page, not just the first. Why React applications fail it, and the four changes that fix most of the failures.

Mar 12, 20268 min read

Architecture

Strangling the Monolith: Zero-Downtime Migration Patterns

How to decouple a monolithic legacy system into event-driven microservices without dropping a single production transaction.

Feb 18, 20268 min read