Atlas

A bilingual (English + Danish) technical dictionary you can explore as a map.

Live site: https://cmaintz.github.io/tech-atlas/

What it is

Atlas explains about 420 technical terms across security, computer science, AI and platform engineering, in English and Danish side by side. Every term is linked to others by typed, sourced relationships ("is a kind of", "requires", "contrasts with" and so on), so the dictionary doubles as a knowledge graph you can navigate.

It is anchored to a real Danish cybersecurity (GRC) course and written first for Danish learners of cybersecurity and governance, risk and compliance who need the English terms and the Danish ones, explained in plain language. It is built to grow into the wider CS, AI and platform domains.

Features

• Dictionary - every term has four facets: a formal definition, a plain-language explanation, an example in practice, and why it matters. Many terms also have a longer technical deep-dive article. Summaries use a closed vocabulary (plain words plus other defined terms), enforced by a lint.
• Explorer - the whole graph in 2D (force, by depth, by time) or 3D, with flowing relationship edges, auto-rotate, domain and relationship filters, a route finder between two terms, and a side panel for the selected term.
• Timeline - terms by the year they entered use, one lane per domain.
• Compare - "don't confuse these" pages for every pair of contrasting terms.
• Study - quizzes generated from the graph plus a hand-written question bank, with spaced repetition to bring terms back when they are due.
• Search - typo-tolerant bilingual search, plus search by meaning (semantic search).
• Accounts and sync - optional sign-in with GitHub or LinkedIn to keep study progress across devices. Everything works without an account.
• Light and dark theme, and every page in English and Danish (/en/, /da/).
• Open data - every term as JSON and CSV, Anki decks, the graph and RSS feeds (see Data).

Screenshots

Term page	Explorer (2D)	Timeline

Tech stack

• Astro 7, static-first, route-based i18n (/en/, /da/)
• Preact islands; Cytoscape.js for the 2D graph and 3d-force-graph for the 3D Explorer
• Tailwind CSS v4; MiniSearch for search
• Content as bilingual YAML validated by Zod (app/src/schema.ts)
• Supabase (auth, progress sync, pgvector) and Cloudflare Workers AI (bge-m3 embeddings) for the optional backend
• Vitest, Prettier, and a custom content lint; gated by Foundry (mise run gate)
• Hosted on GitHub Pages

Architecture in brief

Atlas is a static site. The build reads the YAML terms, derives the graph (depth, inverse edges, visual weight) and renders every page for both languages. All reading, exploring and studying runs in the browser, and progress is kept locally.

An optional backend adds two things on top:

• Accounts and sync through Supabase Auth (GitHub, LinkedIn) with Row Level Security.
• Search by meaning through a Supabase Edge Function that embeds the query with Cloudflare Workers AI (bge-m3) and matches it against pre-computed term vectors.

If the backend is not configured, the site still works; those two features are simply hidden. Setup: docs/SUPABASE_SETUP.md. Security model: docs/SECURITY.md. Reporting a vulnerability: SECURITY.md.

The spec and decisions live in design/ (SPEC.md, adr/, UNIFIED_VISION.md, AUTONOMOUS_DECISIONS.md). When spec and code disagree, the spec wins.

Getting started

Prerequisites: mise (it installs the pinned Node and gitleaks).

mise install                               # pinned tools from mise.toml
cd app
ONNXRUNTIME_NODE_INSTALL=skip npm ci       # skips the ONNX runtime binary download
npm run dev                                # dev server at http://localhost:4321/tech-atlas/

From the repo root:

mise run fix     # format
mise run gate    # lint -> typecheck -> test (incl. the production build) -> audit

A change is done when mise run gate is green. More commands (content lint, graph build, embeddings) are in app/README.md; the working rules for contributors and agents are in AGENTS.md.

Writing content

Terms, articles and quiz questions are YAML files under app/src/content/. The guide is app/content/AUTHORING.md: the four facets, the closed vocabulary, the 12 relationship types and how to write questions.

Every new entry starts as draft: true and shows a "not reviewed yet" banner. A human reviewer checks it and flips the flag. The site's Review page (/en/review/) lists the drafts still waiting, grouped by cluster, with an edit link for each.

Project status

Beta. The site, explorer, study tools and backend are in place. Much of the content was drafted with AI assistance and is being reviewed by hand, entry by entry; until an entry is reviewed it is marked as a draft on the site. Corrections are welcome as issues or pull requests.

Data

The site publishes its content as open data: every term as JSON (/api/terms.json, /api/terms/<folder>/<id>.json), a CSV (/api/terms.csv), Anki import files per language, the derived graph (/graph.json) and an RSS feed of new terms per language. The list, with formats, is on the site's Data page (/en/data/, /da/data/).

License

• Code: MIT, see LICENSE.
• Content (terms, articles, allowed-words lists and the published data): CC BY-SA 4.0, see CONTENT-LICENSE.md for the scope, the attribution line and the reasoning.

Credits

Made by Christoffer Maintz Andersen - GitHub - LinkedIn.