Engineering notes
Software engineering notes about reliability, distributed systems, Java, cloud infrastructure, and AI systems in production.
My blog broke because the public Hashnode GraphQL API stopped behaving like a public read API. I tried the easy recovery paths first, then chose the boring option: move the content into MDX files in the repo and keep going.
In most production incidents involving “missing messages,” the queue is blamed early. SQS is down. The message was dropped. AWS lost it. True message loss inside managed queue infrastructure is extremely rare; most incidents come from visibility timeouts, retries, consumers, redrive policy, or observability gaps.
Modern backend services rarely perform isolated work. A single request often fans out into multiple network calls, database queries and asynchronous operations. The service is effectively coordinating work under a deadline, and Java 21 gives teams better primitives for that coordination.
Large language model APIs feel deceptively simple from an engineering perspective. You send a prompt, you receive text. Compared to provisioning databases, tuning JVM memory or debugging distributed locks, the interface feels almost too easy.
Over the last year, many products added AI features: chat assistants, automatic summaries, classification, recommendations, drafting emails, generating documentation, and suggesting actions. In many cases, those features changed the reliability surface of the product.
Modern backend engineering increasingly revolves around distributed systems. As a consequence, many technical interviews, even for senior and leadership roles, are designed around deceptively simple prompts that test tradeoffs rather than memorized architecture diagrams.