Writing
Engineering notes from the work.
Focused accounts of MCP resilience, evidence-grounded AI, search architecture, debugging, test infrastructure, and dependable production delivery.
04 / Writing
Notes from debugging and delivery.
Incident Context: tracing production failures without guessing
Building an incident-investigation agent that follows structured evidence paths, separates facts from inference, and fails closed when its knowledge source is unavailable.
I Injected the Same Failures Into 5 MCP SDKs. Here’s What I Found.
A cross-SDK comparison of timeout, cancellation, malformed-response, and transport-failure behavior under the same deterministic scenarios.
What Happens When an MCP Tool Call Never Returns?
Why a hanging tool differs from a slow one, what a timeout actually proves, and how to verify client recovery and server state afterward.
How I Designed a Production Global Search System on PostgreSQL
Designing a parser and AST for nested queries, then reducing measured search latency from roughly 40ms to 12ms with PostgreSQL GIN indexing.
Fixing an Infinite Loading State in the npmx Code Browser
Tracing an idle request state and restoring the oversized-file fallback in a Nuxt code browser.
Beyond Test Automation
Designing deterministic execution, parallel CI, reusable infrastructure, and release-focused coverage.