Articles
Field notes for teams shipping serious software.
Practical notes from the work of making systems clearer, safer, and easier to operate.
ct
Scraping pipelines that survive anti-bot churn
Recurring data collection breaks quietly when storefronts change and bot walls tighten. Production scraping needs matching logic, freshness budgets, and alerting long before the dashboard goes stale.

What AI answer engines actually read — an exposure checklist for B2B sites
GPTBot, ClaudeBot, PerplexityBot, and OAI-SearchBot do not read your site the way Google crawls it. Teams that want citations need explicit crawler policy, machine-readable summaries, and answer-shaped content.

Evaluation harnesses for LLM features you can actually ship
Demos impress; labeled datasets, regression suites, and rollback criteria get you through security review. We treat prompts like deployable code—with tests, owners, and release notes.

Incident response for teams without a 24/7 NOC
You do not need a glass bridge to run sober incidents—you need roles, comms templates, and a blameless timeline that fits your time zones and languages.

When internal tools earn a roadmap, not a side channel
Spreadsheets and shadow IT stick around when leadership never funds internal products like products. We score tool requests the same way we score customer features—impact, risk, and cost of delay.

RTL and Arabic SEO — where mechanical issues torpedo migrations
Hreflang, canonical discipline, and Core Web Vitals do not fix themselves when English is mirrored into Arabic. Teams that skip the mechanical layer waste translation budget on pages search engines never ranked fairly.

Designing APIs that survive your first enterprise pilot
Versioning, idempotency, and honest error models matter long before a logo lands on the homepage. We treat external APIs like contracts finance can audit.

Ship a fast marketing site without babysitting servers
Marketing and documentation sites should load instantly, stay secure by default, and deploy on every merge with a modern static-first delivery pipeline.

Designing internal mobile workflows crews actually adopt
Field teams do not reject mobile tools because they dislike technology—they reject tools that ignore offline time, clumsy authentication, and the reality of gloves, sunlight, and partial attention.

Practical AI starts with grounded retrieval, not theatrical demos
Executive-friendly GPT demos rarely survive contact with messy operational data unless retrieval, citations, and evaluation harnesses are designed like any other production service.

Dashboard mania will hide sloppy upstream workflows
A vivid chart rarely fixes stale source data or unclear ownership. We rebuild dashboards only after the underlying events, definitions, and SLAs are honest.

Security basics every custom software initiative should budget
Threat models are not academic exercises—they decide whether your first big customer will accept your DPA or delay procurement until a competitor proves control.