designfuchen/word-atlas

Worth a look if you are studying for IELTS or want a ready-made vocabulary dataset; skip it if you need audited definitions or a team study platform.

The problem

A learner opens a vocabulary list, scrolls to abandon, and never gets a reason to return. The gap is not exposure but retrieval: a flat list gives no cue, no schedule, and no signal of what was actually learned. Word Atlas replaces the list with one card per word — an illustration, UK/US audio, example sentences and a short quiz — and schedules the next appearance with FSRS. The result is a repeatable daily loop rather than a one-pass reading session.

What's different

The distinguishing piece is the asset set: the project says all 7,298 words have an original illustration and that the corpus contains 14,596 example sentences and 22,905 recordings, with the illustrations keyed to the first example sentence. Keeping that coverage consistent is a content problem, not a UI one, and the repository exposes both the vocabulary JSON and CSV plus the image and audio directories, so the material can be reused outside the app: an Anki deck, a small quiz tool or a test set.

The review layer is FSRS (the same family of algorithm used by newer Anki versions), with four question types and a rule that only the first attempt counts, so retries do not inflate accuracy. The star map renders the whole vocabulary in WebGL2, with a fallback for unsupported devices, and progress can be exported to a file and moved between devices. Offline use is limited to what has already been cached, not the full 7,298-item library. The trade-off is provenance: the author states in the README that definitions, sentences and illustrations were generated with language and image models plus ElevenLabs speech, checked by automated structure and consistency tests but not reviewed entry by entry. Treat it as a study aid, not an authority on meaning or usage.

Who should try it

A good fit is an IELTS candidate who wants a free, no-account daily review habit and who values visual cues over a bare word list, or a developer who wants a licensed vocabulary dataset to build a deck or prototype around. A poor fit is someone who needs human-verified dictionary content, a teacher-managed classroom with shared progress, or a full offline download of all words and media. If the aesthetic and sound design feel distracting rather than motivating, the app can be opened in the browser with no commitment.

Requirements and limitations

The app runs in a browser at the hosted demo; self-hosting needs Node.js 20+, and the build output is a static site that can go on any static host, including under a subpath. The dataset itself is downloadable as JSON or CSV, with field notes in the dataset documentation, so much of the value is available without running the app.

The important ongoing cost is trust: because content is auto-generated and not line-by-line reviewed, a wrong sense or an odd illustration must be reported or corrected by a contributor, and the project welcomes issues and PRs. The README also says this is not an official IELTS list and does not promise a score improvement. If accuracy matters for your use, budget time to spot-check entries in your domain before relying on them.

HotRepo's take

HotRepo's only dated record is a 2026-10-25 observation of 309 total stars and #24 on the NEW-REPOSITORY board, not an overall Trending rank; there is no comparable daily-board history, so persistence cannot yet be judged. The dataset is the durable artifact here: check the content license and the per-entry fields before committing a production pipeline to it, and decide whether an automatically generated vocabulary fits your accuracy bar.