QamoosTech (قاموس تك)

PROBLEM

Arabic-speaking engineers spend their careers in English — documentation, code reviews, client emails, stand-ups, interviews — and most glossaries either translate word-for-word or ignore how the words are actually used. QamoosTech explains each term in plain Modern Standard Arabic and in English, with where you hear it, real workplace examples, the common mistake, and how to say it.

WHAT'S IN IT

• +Arabic-first, fully right-to-left site with a language switcher built so a third language is a folder, not a rewrite.
• +Instant search that normalises Arabic (diacritics, alef variants) and matches English and Arabic in one box.
• +An animated hero — drifting aurora, floating bilingual term cards with pointer parallax, a rotating clickable headline — that respects reduced-motion and still scores 93 on mobile Lighthouse performance.
• +A "Surprise me" discovery loop on every page, plus previous/next and related terms, so one lookup turns into ten.
• +Pronunciation audio for 277 of the 727 terms so far, generated with Gemini TTS across two API accounts by a resumable, quota-aware script, with the browser voice as the fallback.
• +A shared model pool that rotates API keys and models across tiers with fallback and daily budgets, used for drafting and for an automated Arabic/English quality review.
• +Three workplace examples for every term in both languages, checked by the content validator, which fails the build when the languages drift apart or a related term points nowhere.
• +A tag filter across 11 tags and flashcard practice, alongside instant search and categories.
• +A number-sync script keeps the published figures in step with the content, so the counts cannot drift.
• +Open source from day one: MIT for the code, CC BY-SA 4.0 for the content, with a contributor guide.

ROLE

Solo builder — content architecture and authoring pipeline, the Next.js site and design system, search, audio generation, SEO, and deployment.

NOTABLE DECISION

Made the content the product and kept the code generic: one Markdown file per term per language, with a validator that fails the build if the languages drift apart, a related-term id is wrong, or the wrong writing system sneaks into a file. Adding a new language is a folder and a few UI strings, nothing else. That discipline paid for itself when AI-drafted Arabic arrived with Hindi and Thai characters inside it — the scan caught what a model review missed.