Mindvalley Stories
Case Study · AI Mastery

Someone who had never opened a terminal put Estonian inside macOS

Four Estonian dictionaries with 406,725 entries now live inside Apple's own Dictionary app, built in under half a day of hands-on time.

This builder had never opened a terminal. Working with Claude Code in Estonian, they made their language a native part of macOS: select a word anywhere, press a shortcut, and read a real dictionary entry. It is a striking proof that careful thinking about what to check can matter more than knowing how to code.

Hebe Arus
A non-programmer and Estonian speaker who had never written code or used a terminal before this project.
Estonian dictionaries and spellcheck for macOS

Before

Before this project they had never opened a terminal. On their Mac, Estonian had no place in the system: no built-in dictionary lookup and no Estonian spellcheck, which macOS does not offer. With 14 grammatical cases, Estonian words rarely appear in their dictionary form, so even a plain word list would not have been enough.

“Before this project I had never opened a terminal.”

In their own words

The turning point

They used Claude Code as their entire development environment, working in Estonian. Since they could not review the code, they decided to review the results instead: entry counts, coverage percentages and entries for words they picked themselves. They validated generated files against Apple's schema before building, and tested with deliberate typos rather than hope.

“I could not review the code, so I reviewed the results: entry counts, coverage percentages, and entries for words I picked myself.”

In their own words

After

Across six calendar days and under half a day of hands-on time, they produced a 1.5 GB offline toolkit. It holds four dictionaries running inside Apple's Dictionary app, links 5,168,360 inflected forms back to their headwords, and teaches 5,328,332 word forms to the system spellchecker. They also ported it to iPhone and wrapped it in a one-click installer.

What it lets people do

An Estonian speaker can select any Estonian word anywhere in macOS, press Ctrl+Cmd+D and read a real entry, even when the word appears in an inflected form. A non-technical person can install it without a terminal, and Estonian spellcheck now works on the system.

What they learned

They found that the hard part is not writing code but knowing what to verify. AI can be confidently wrong about how an operating system works, and a quiet success can hide a total failure, as when the spellchecker looked perfect while checking nothing. They now treat documented dead ends as deliverables too.

“The hard part is not writing code. It is knowing what to verify.”

In their own words

See the work

Estonian-English entry for raamat inside Apple's Dictionary app
System-wide lookup popup showing dictionaries side by side
Inflected form raamatutega leads to raamat with full case table

Estonian dictionaries and spellcheck for macOS was built during AI Mastery.

Read the full submission
Before this project I had never opened a terminal. Total hands-on time was under half a day, spread across six calendar days — most of the elapsed time was an unattended download. What came out is a 1.5 GB offline toolkit that puts Estonian inside macOS: select any Estonian word anywhere in the OS, press Ctrl+Cmd+D, and read a real dictionary entry. ACHIEVEMENTS - Four dictionaries, 406,725 entries, running inside Apple's own Dictionary app — not a separate application, but the OS feature itself. - 5,168,360 inflected forms map back to their headwords. Estonian has 14 cases, so without this a dictionary is unusable: you almost never meet a word in its dictionary form. - Estonian spellcheck, which macOS does not offer: 5,328,332 word forms taught to the system spelling service. - Also ported to iPhone, and packaged as a one-click installer so a non-technical Estonian can install it without a terminal. - Coverage measured against a real frequency list, not claimed: 92% translation and 98% inflection coverage on the 1,000 most common words. TOOLS Claude Code (Opus) as the entire development environment, working in Estonian. Python, SQLite, Swift, Bash — ~2,744 lines. Ekilex API (Institute of the Estonian Language), Apple Dictionary Development Kit, pyglossary, Cocoa NSSpellChecker. METHODOLOGY - I could not review the code, so I reviewed the results: entry counts, coverage percentages, and entries for words I picked myself. - Validate before assembling. 178 MB of generated XML is checked against Apple's schema before the build tool sees it — which is how broken tags in national open data got caught. - Test with deliberate typos, not with hope. LEARNINGS - The hard part is not writing code. It is knowing what to verify. - AI is confidently wrong about operating system internals. My commits read "macOS 26 DOES support Hunspell" and, two commits later, "the Hunspell path is a dead end." A configuration entry proves configuration, not function. - Silent success is the dangerous failure mode. The spellchecker looked perfect — red underlines gone — while macOS was classifying Estonian as Finnish and checking nothing at all. - Negative results are deliverables. Three impossible approaches are now documented with evidence, so the next person loses hours instead of weeks. RELEVANT URLS - Ekilex, Institute of the Estonian Language: https://ekilex.ee - EKI open data archive: https://arhiiv.eki.ee/litsents/ - Estonian wordlist (160k words, 5.4M forms): https://github.com/KristjanPikhof/Estonian-Wordlist-Enriched-Ekilex Source data used under CC BY 4.0, CC BY-SA 4.0 and Public Domain, with attribution embedded in every dictionary bundle.

Written 10 August 2026 Submitted by the storyteller

Every story is written by a Mindvalley student or by someone our teachers have worked with. How we collect and check stories

More from AI Mastery

Estonian dictionaries and spellcheck for macOS: a AI Mastery case study | Mindvalley Stories