Hebrew Tech Vocab List
July 28, 2026 · View on GitHub
Project
For those who don't know me, I'm an Irish-born immigrant who moved to Israel ..... squirms in embarrassment at state of Hebrew .... a while.
I thought I would create a repo to gather some translations for niche tech words. This is an updated take on an old Anki board project that I created in that dinosaur era that we lived through before the advent of AI.
The "reasoning": Google Translate (etc) are great but, from long experience, there's often a gulf between what they provide and ... what people actually say.
Also:
Hebrew, like all languages, is in a constant state of flux. Tech vocab, in any language, is often new - and tends to propagate from English to international equivalents.
Sometimes, there's both a more official word (often English or a direct transliteration like בלוטוס) and a word that is a more "correct" version that falls out of favor.
Scope: wider than software
Ten years in, day-to-day Hebrew is fine. What isn't fine is the specialised vocabulary you meet doing up an apartment — and a lot of it is specialised even for Israelis. Standing in front of a supplier's site with eight cable categories, every one of which translates cleanly and none of which tells you which to click, is the situation this repo is for.
So "tech" here means technical, not just software: electrical, plumbing, hardware, tools, construction, and the counter Hebrew for actually buying the stuff. The failure mode is identical in all of them — the dictionary gives you a word that is correct and useless.
The knowledge already accumulates, one small purchase at a time, in my head. The point of the repo is to get it out of there and into something with spaced repetition attached.
Method
Two passes, and the second one is where the value is.
Pass 1 — machine. A screenshot goes in: a product listing, a category grid, a
shelf label, an invoice. Claude reads every Hebrew string on it, decides which
terms earn an entry, researches the register rather than just the meaning, and
writes entries with human_verified: false and an honest confidence rating.
Low confidence is a feature — it's the review queue.
Pass 2 — human. I periodically attempt to bribe native Hebrew speakers with
beer, or appeal to their charitable instincts, to review the file. Confirmed
entries flip to human_verified: true. Anything they say about usage — "nobody
says that any more", "that's builder Hebrew, not what you'd say in a shop" — goes
into human_notes verbatim. That commentary can't be regenerated by anyone.
There's a third input that skips pass 1 entirely: things I work out myself in the field. A tradesperson corrects me, or a purchase goes wrong because I asked for the wrong thing. Those go straight in as verified, with the story attached, because the story is the mnemonic.
All of it runs through the vocab-capture skill — paste an image, say whether
you know the term or want it researched, and it handles the rest.
dictionary/dictionary.json is the source of truth.
What's in here
| Path | |
|---|---|
dictionary/dictionary.json | The dictionary. Single source of truth. |
dictionary/schema.json | Field semantics, including what register means and why he holds the useful word rather than the correct one. |
sources/ | Screenshots, kept as reference, one folder and one markdown note per source. |
process/to-add/ | Words wanted, where there's no source to capture. |
scripts/ | validate.py, stats.py, export_anki.py. Stdlib only. |
anki/ | Generated deck. Regenerated wholesale — never edit. |
.claude/skills/vocab-capture/ | The capture workflow. |
.claude/commands/ | /verify /find /stats /export-anki /tts-script |
Entries record register, not just meaning
The single design decision everything else follows from: he is the word to
actually use, not the word that is officially correct. Where those differ, the
official one goes in alternatives with a note saying who uses it.
The circuit breaker entry is the canonical example. The Academy term is מפסק אוטומטי זעיר, abbreviated מא"ז on the packaging and on your electrician's invoice. Say that in a shop and you'll get a blank look. Everyone says אוטומט. Google Translate will not tell you that, and it is the only thing you needed to know.
Nouns carry gender and plural, because a dictionary you can't build a sentence from is a word list. Examples are sentences you'd really say — קפץ לי האוטומט, not המפסק האוטומטי נותק.
Sister repo: the conventions
A word is often only half of what you needed. Israel-Procurement-Morsels holds the other half — the unwritten rules that the vocabulary unlocks.
Knowing גליל means "coil" belongs here. Knowing that searching גליל is the
only way to reach the whole-reel listing, which is 17% cheaper than cut-to-length
belongs there. The first is a translation; the second is what actually costs you
money if nobody tells you.
Entries here link out to the morsel when there is one, and vice versa.
Learning resources
To facilitate spaced repetition, the dictionary is periodically used as the basis
for a TTS-generated "episode" for vocab learning and recap — /tts-script.
Requires English TTS, Hebrew TTS or multilingual TTS; see
Hebrew-TTS-Providers
for what currently handles Israeli-accented Hebrew with correct stress.
There's also an Anki export, which is where this project started, twice.
On the screenshots
Product listings are saved into sources/ as reference for a personal glossary,
unmodified, with the vendor named. Some are annotated — the circles are my own
reading of the page and are themselves part of the record, so annotated versions
sit alongside the originals rather than replacing them.