Deploying the still-mock-backed app to Cloud Run gated two ways (Slack SSO at the app layer and an office-IP allowlist) with a standalone container and keyless GitHub Actions via Workload Identity Federation. Plus the cost call between an app-layer IP check and Cloud Armor, and the auth gate that forced collapsing an internal self-fetch.
A deep dive into why upgrading from Node 20 to 24 resulted in lower CPU usage, fewer GC pauses, and healthier event loop metrics. Covers V8 13.6, Maglev JIT, smarter GC heuristics, and more.
What broke when upgrading Node.js from 20 to 24 in a production Next.js app running in Docker, and how I fixed each issue: npm@11 breaking changes, ignore-scripts, git auth in multi-stage builds.
How I fixed the bot serving outdated knowledge after GCS file updates by adding a reload endpoint, OIDC authentication, and wiring up GCS change notifications through Pub/Sub.
Replacing an in-memory mock with local Postgres: Docker, a Drizzle schema and first migration, seeding, swapping the data layer, and designing for private-network cloud access.
A form sends a URL. The bot decides whether the page lives in a repo it can change, and if so it triggers a CI workflow where an AI coding agent makes a fixed, repetitive edit and opens a pull request, then reports back into the Slack thread. Covers classifying a page via the GitHub contents API, the reusable-workflow pattern, an AI agent in CI, and a secured callback.
A report form sends a URL. The bot normalizes the URL, looks it up in a master spreadsheet to find who owns that page, and @mentions them in-thread, with sensible fallbacks. Covers URL normalization, an efficient bottom-up sheet lookup, and Slack mention resolution including subteam groups.
Parsing a multi-field Slack Workflow form into structured data, appending it to a Google Sheet, posting a review message with an Approve button, and on approval booking a Google Calendar event. Covers form parsing, the Sheets API, Slack interactive buttons, and the Calendar API.
Concrete gotchas from building the UI of a status-driven workflow tool on Next.js 16 and React 19: intercepting + parallel routes for modals, the React Hook Form + Zod typing trap with default values, gating a stepper by status and role, and pinning a gantt with the one legitimate inline-style exception.
My hobby recipe app got paused after 90 days of inactivity and could no longer be restored from the dashboard. Here is how I rescued the data from a backup and rebuilt the stack on platforms that do not freeze - swapping Supabase for Neon + Drizzle, Auth.js, and Vercel Blob.
Upgrading Swiper from v8 to v11 seems harmless until your carousels start thrashing the page. A deep dive into why loop mode broke, what loopAddBlankSlides actually does (nothing useful), and how to fix it with a wrapper component.
Why I replaced prism-react-renderer with shiki for code highlighting in my Next.js blog, and how the migration went.
A new design draft and a fresh requirements list arrived mid-build. How I triaged them against priority, deferred the integration-heavy features without stalling, kept the plan doc as the source of truth, and decided the schema was finally stable enough to stand up the database.
Why I built the entire UI of an internal workflow tool before standing up Postgres, and how the mock layer worked: an in-memory store, MSW interception, Zod as the single source of truth, and a cookie-based mock session to exercise every role.
Configuring AGENTS.md, code conventions, and GitHub Copilot review rules before the codebase exists. The patterns that make AI coding assistants useful instead of a slow drag.
Walking through the cloud, database, framework, and surrounding choices for a small internal tool on GCP. Why Cloud Run instead of GKE, why Postgres instead of NoSQL, why skip Redis, and the rest of the decisions worth thinking through before writing code.
A deep dive into how JVM handles memory management through garbage collection, its types, and best practices.
Designing monitors that detect only your service’s own failures, filtering out upstream noise.
Refined Datadog queries for ALB metrics, latency, unit handling, and rate computations: mistakes learned and correct patterns.
