AMIRA — Africa Multiple Interactive Research Atlas

Interactive research atlas for the Africa Multiple Cluster of Excellence, exposing the cluster's WissKI/MongoDB archive as a browsable, analysable, visually-rich web app. Built with SvelteKit 5, ECharts 6, MapLibre GL 5, and Tailwind CSS 4 — shipped as an installable static PWA to GitHub Pages.

Live: https://amira.africamultiple.uni-bayreuth.de/

Scope

• 3,975 research items across 92 projects
• 1,394 persons, 492 institutions, 84 groups
• 146 cluster publications harvested from ERef Bayreuth and EPub Bayreuth, deduplicated across the two repositories
• 13 research sections — 6 Phase 1 (Affiliations, Arts & Aesthetics, Knowledges, Learning, Mobilities, Moralities), 6 Phase 2 (Accumulation, Digitalities, Ecologies, In/securities, Re:membering, Translating), and an "External" pseudo-section
• 4 partner universities (Bayreuth lead + UNILAG, UJKZ, Moi, Rhodes) plus 1 privileged partner (UFBA's CEAO, Salvador de Bahia)
• 2 external collections (BayGlo2025, ILAM)
• Per-entity detail dashboards for languages, subjects, tags, people, institutions, genres, resource types, groups, locations, research sections, and projects — all backed by precomputed JSON dumps under static/data/entity_dashboards/

Features

Dashboard

• Overview — summary stat cards (items, projects, contributors, institutions, languages, countries, subjects/tags), a cluster-locations MiniMap showing the geographic reach of the partnership, a featured-collections sneak-peek grid, global filter panel (resource type, language, university), stacked timeline by type, resource-type pie, subject bar chart, tag word cloud, research-section distribution, and a Research Section × University heatmap matrix. Phase 1 sections chart separately from Phase 2 (which begins June 2026).
• What's New — recent additions within a selectable window (3 / 6 / 12 months): top projects by new-item count, list of the most recent items, and a quick pivot into the detail views.

Research

• Research Sections — the 13 sections with full descriptions, objectives, work programmes, spokespeople / PIs, members, and project Gantt timelines. Phase 1 sections show project charts; Phase 2 sections surface spokesperson info only (no projects until June 2026). Each section also opens an EntityDashboard of stacked timeline, resource type / language / subject breakdowns, decade heatmap, subject trends, time-aware chord, contributor network, and origin → current geo flows.
• Projects — faceted browser (research section, institution) with detail views showing description, PIs, members, institutions, paginated items, and WissKI links. Overview adds a Gantt timeline, beeswarm (projects by section and year, sized by item count), and bar charts for sections and institutions. Per-project detail pages now ship the full Phase 4 dashboard: stacked timeline, type / language / role pies, subjects bar, word cloud, type × decade heatmap, subject trends, sunburst, chord + time-aware chord, sankey, contributor network, geo flows, locations map, and knowledge graph.
• Research Items — full-text search and collapsible facets (subject, tag, country, project, language, resource type) over all 3,975 items, with a sortable table view. Detail view exposes title, abstract, contributors (role-qualified persons / institutions / groups), subjects (LCSH), tags, origin locations on a MiniMap, dates, language, identifiers, project, and a per-item entity knowledge graph with fullscreen mode. Below the detail card you'll find a SiblingItemsSparkline (year-by-year project timeline with the current item highlighted) and a SimilarItemsStrip (top-8 semantic-kNN matches via Gemini embeddings).
• Publications — cluster bibliography synced from ERef Bayreuth (EXC 2052 project view) and EPub Bayreuth (Africa Multiple division view). The fetcher dedupes by DOI / ISBN / fuzzy title+year+author and merges cross-listed records; each card shows DOI, ERef, and / or EPub buttons depending on which repos carry the item. Stat cards (total, types, languages, matched authors), per-year stacked-by-type timeline, language pie, keyword wordcloud, full-text search across title / abstract / journal / author / DOI / keyword, faceted filters (type, year, language, keyword), and per-item BibTeX / RIS export plus bulk export of any filtered subset.

Collections

• Featured Collections — curated showcase of photography and multimedia collections. Detail pages offer three synchronised views of the same deduped item set:
  • Masonry — responsive photo grid that reshuffles on resize and opens items in a photo lightbox (keyboard nav, metadata sidebar, deep-link to WissKI).
  • Map — MapLibre GL cluster map of capture locations with SPA-routed popups.
  • Timeline — chronological stacked view with zoom and type breakdown.
  • Faceted filters (creator, subject, tag, country, year), WebP thumbnails shipped with the build for fast first paint, and deduped alias counts for photos that appear in multiple records.
  • History-aware navigation — tab switches push to browser history and the open photo is mirrored to the URL (?photo=…), so browser Back closes the lightbox and steps through tab changes instead of jumping to the collections list.

Directory and Categories — unified card-based browse + per-entity dashboards

All directory and category pages (People, Groups, Institutions, Genres, Languages, Locations, Resource Types, Subjects & Tags) share a consistent card-grid layout backed by reusable components (EntityCard, EntityBrowseGrid, EntityToolbar, EntityDetailHeader, EntityItemsCard) and utilities (applyEntitySort). Each card shows the entity name, descriptor, icon, count with label ("items", "projects", etc.), and type-specific meta chips; pages use a single toolbar with full-text search, count / alphabetical sort toggles, and a results badge. The grid paginates at 48 cards per page to keep long lists (1,000+ subjects, tags, cities) fast.

Detail views are opened via URL query params (e.g. /people?name=John%20Doe, /genres?genre=Broadcast) and the selection is driven by a writable $derived over $page.url.searchParams — so browser Back automatically clears the detail view and restores the card grid, deep links are shareable, and reload preserves state.

Each entity detail page now mounts an <EntityDashboard> populated from a precomputed per-entity JSON file. Layouts are declared in entityDashboardLayouts.ts; slots auto-hide when their payload is empty, so the grid never carries placeholder cards.

Directory

• People — affiliation filter (searchable combobox), role filter, "hide empty" toggle, and rich profile views with research sections, projects (as PI or member), affiliations, a per-person dashboard (stacked timeline, resource types / languages / roles, top subjects, co-credited persons chord, projects + co-contributors network, affiliated institutions network, locations), paginated items with a per-item role badge, and a collaboration & influence knowledge graph.
• Groups — research groups and organisational units that author items, with per-group dashboards (timeline, types, languages, subjects, members & collaborators, locations) and knowledge graph.
• Institutions — partner / contributor filter toggle, per-institution dashboard (stacked timeline, types / languages, top subjects, affiliated contributors, contributor + affiliation networks, locations), projects (with PIs, dates, sections), people, items, and knowledge graph.

Categories

• Genres — MARC genre classification with a top-20 bar chart, a genre × language heatmap, and a card grid; per-genre detail pages include the standard EntityDashboard (timeline, types / languages, subjects, contributors, locations).
• Languages — ISO 639-2/3 codes rendered as full English names throughout. List page adds a stacked-area timeline and a language × resource-type heatmap; per-language detail pages render a 9-card dashboard plus a country choropleth.
• Locations — country / region / city browser with a MapLibre GL browse map on top of the list (markers coloured by type, clickable popups that SPA-route into the detail page), a country choropleth, top-cities bar, and tagged-items timeline. Each location detail shows a self-location MiniMap, the standard dashboard, and the knowledge graph.
• Resource Types — text, sound recording, still image, moving image, cartographic, mixed material, etc., with pie and bar charts, a stacked-area timeline, and a type × language heatmap; per-type detail pages render a full dashboard with subject trends.
• Subjects & Tags — toggle between LCSH controlled vocabulary and free-form tags, unified tag icon across both views, animated word cloud (up to 200 terms, adjustable size) with click-through, plus subject-trends overlay and an LCSH-vs-tag pie. Per-subject / per-tag detail pages add a co-subject network and the standard dashboard.

Visualize

• Project Explorer — cross-project analytical workspace with section / project selectors and synchronised charts.
• Compare — single page (/compare/[type]) with a tab bar that switches among projects, people, institutions, subjects, languages, and genres. Every type renders the same primitives — stat row, 6-axis profile radar, shared-subjects chips, side-by-side stacked timelines + resource-type pies + top-subjects bars + language bars + contributor bars — with the data source swapped per type. /compare/projects covers university × project filters via on-the-fly aggregation off the loaded collections; the other types load precomputed entity-dashboard JSON. The legacy /compare-projects URL still works as a redirect to /compare/projects.
• Network — five interactive force-directed graph tabs, all weighted (edge thickness = strength of tie) and community-coloured:
  • Contributors ↔ Projects — weighted bipartite graph; edge width equals the number of items a person contributed to that project
  • Co-authorship — person ↔ person projection of the contributor graph, edges for pairs with ≥ 2 shared items
  • People ↔ Institutions — bipartite affiliations view
  • Institution collaborations — institutions joined when they share people on the same project
  • Discursive communities — Louvain communities detected across the whole archive, with the top-PageRank anchor of each cluster, built from the precomputed _meta.json
  • Filters: university, max nodes, resource type
• Semantic Map — 2D UMAP projection of Gemini-embedded item vectors. Scatter colourable by university, resource type, language, or research section; searchable and filterable; click a point to see the item and its top-12 cosine-similar neighbours. Similar-items payload is lazy-loaded on first selection.

Cross-linking, WissKI, PWA, and SEO

Every named entity is a link:

• PIs, members, and contributors → People (persons) or Institutions / Groups (by qualifier)
• Project names → project detail
• Item titles → item detail + knowledge graph
• Section badges → section detail
• Institution badges → institutions page
• Language badges → languages page
• Location origins (city / region / country) → locations page
• Resource-type badges → resource types page
• Subject and tag badges → subjects & tags page

WissKI Navigate — Optional deep-links to WissKI entities surface throughout the dashboard via pre-computed URL mappings (dev.wisski_urls.*.json), connecting every dashboard record back to its source in the WissKI knowledge base.

Chart downloads — Every ECharts visualisation exposes a download button in its card header that exports the chart as a PNG with the card title, subtitle, and export date composited on top, using the site's typography and theme (light or dark).

PWA — The site ships a Web App Manifest (static/manifest.json) and a service worker (src/service-worker.js) so it installs as a standalone app on desktop and mobile, with offline shell and shortcuts to What's New / Projects / Research Items / Network. Theme colour, maskable icons, and orientation hints are all set.

SEO — Every route ships a <SEO> component that emits Open Graph + Twitter Card tags and a canonical URL; the build also pre-renders /sitemap.xml covering both static routes and every entity-detail URL discovered by scanning static/data/entity_dashboards/.

Mobile UX — The whole app is touch-friendly on a phone: pagination controls scroll the list back into view on tap, stat cards stack with sensible truncation, the Gantt timeline collapses to a vertical list, empty states render at full card width, and the sidebar slides over the content rather than reflowing it.

Tech Stack

• Framework — SvelteKit 5 with Svelte 5 runes ($state, $derived, $derived.by, $effect, $props)
• Build — Vite 8 with Rolldown
• Charts — ECharts 6 + echarts-wordcloud
• Maps — MapLibre GL 5 (LocationMap, MiniMap, ChoroplethMap, GeoFlowMap, PhotoMap — all with a Flat / Globe projection toggle where applicable)
• Embeddings — Gemini Embedding 2 (gemini-embedding-2-preview, 768-dim) projected to 2D with UMAP for the semantic map
• Styling — Tailwind CSS 4 via @tailwindcss/vite, HSL CSS-variable theming, dark mode
• UI Components — custom shadcn-svelte-inspired primitives (Card, Badge, Pagination, StatCard, ChartCard, Tabs, Select, Combobox, ScrollableTable, EmptyState, SEO…)
• Icons — @lucide/svelte
• Linting — ESLint 10 + typescript-eslint + eslint-plugin-svelte
• Formatting — Prettier 3 + prettier-plugin-svelte
• Deployment — GitHub Pages via @sveltejs/adapter-static and GitHub Actions

Chart Components

Development

npm install
npm run dev           # dev server
npm run build         # production build
npm run preview       # preview production build
npm run check         # svelte-kit sync + svelte-check
npm run lint          # ESLint
npm run format        # Prettier write
npm run format:check  # Prettier check (CI uses this)

CI runs format:check, so run npm run format locally before committing if the check fails.

Project Structure

src/
├── lib/
│   ├── components/
│   │   ├── ui/              # Reusable UI primitives
│   │   │   ├── card*.svelte, badge, button, input, select, combobox, tabs
│   │   │   ├── pagination, stat-card, chart-card, scrollable-table
│   │   │   ├── collection-item-row, back-to-list, empty-state
│   │   │   ├── section-badge, wisski-link, scroll-to-top, seo
│   │   ├── charts/          # ECharts + MapLibre chart components
│   │   │   ├── EChart.svelte              # Base wrapper
│   │   │   ├── ChartDownloadButton.svelte # PNG export with title / subtitle composited
│   │   │   ├── chart-registry.ts          # Context bridge: chart instance ↔ ChartCard header
│   │   │   ├── Timeline, StackedTimeline, StackedAreaChart
│   │   │   ├── BarChart, PieChart, WordCloud
│   │   │   ├── HeatmapChart, CalendarHeatmap, BeeswarmChart, BoxPlot
│   │   │   ├── RadarChart, GanttChart
│   │   │   ├── SankeyChart, SunburstChart, TreemapChart
│   │   │   ├── ChordDiagram, TimeAwareChord
│   │   │   ├── NetworkGraph, ContributorNetwork, EntityKnowledgeGraph
│   │   │   ├── SemanticScatter
│   │   │   ├── LocationMap, MiniMap, ChoroplethMap, GeoFlowMap, LocationsMapView
│   │   │   ├── map/                       # Projection toggle, marker / popup builders
│   │   │   └── utils/                     # Shared option builders & tooltip formatters
│   │   ├── dashboards/      # Per-entity detail dashboards
│   │   │   ├── EntityDashboard.svelte         # Generic chart-grid renderer
│   │   │   ├── EntityDashboardSection.svelte  # Loader wrapper for pages
│   │   │   ├── ChartSlot.svelte               # Dispatches chart key → component
│   │   │   └── entityDashboardLayouts.ts      # ENTITY_LAYOUTS + ChartKey + emptiness rules
│   │   ├── compare/         # Side-by-side compare primitives
│   │   │   ├── CompareTabs, CompareStatRow, CompareSharedSubjects
│   │   │   ├── CompareProfileRadar, ComparePair
│   │   │   ├── ProjectsCompare (on-the-fly aggregation)
│   │   │   ├── EntityCompare (precomputed JSON)
│   │   │   └── compareTypes.ts, compareProfile.ts
│   │   ├── collections/     # Featured-collection components
│   │   │   ├── CollectionHeader, CollectionIndexCard
│   │   │   ├── PhotoCard, PhotoFacets, PhotoLightbox
│   │   │   ├── PhotoMasonry, PhotoMap, PhotoTimeline
│   │   │   ├── ViewModeTabs, photoHelpers.ts
│   │   ├── entity-browse/   # Unified card-grid components for directory / categories
│   │   │   ├── EntityCard, EntityCardGrid, EntityBrowseGrid (grid + pagination)
│   │   │   ├── EntityToolbar (search + sort + total), EntityDetailHeader
│   │   │   ├── EntityItemsCard, SearchableItemsCard
│   │   │   ├── EntityEmptyHint, sort.ts (applyEntitySort)
│   │   ├── layout/          # Sidebar (grouped nav), Header, FilterPanel
│   │   └── research-items/  # ItemDetail, ItemFilters, ItemTable, itemHelpers,
│   │       │                # SiblingItemsSparkline, SimilarItemsStrip
│   ├── stores/
│   │   ├── data.ts          # Raw data + derived stores (projects, persons, collections…)
│   │   └── filters.ts       # Global filter state
│   ├── styles/              # Tokens, animations, component CSS, sidebar, maplibre
│   ├── types/               # TS interfaces (domain, collection, charts, geo, mongo, embeddings, category-index)
│   └── utils/
│       ├── transforms/      # dates, grouping, extractors, network, charts, filters
│       ├── loaders/         # mongoJSON, collectionLoader, geolocLoader, embeddingsLoader,
│       │                    # entityDashboardLoader (per-entity JSON + manifest)
│       ├── external.ts      # Virtual external projects (BayGlo2025, ILAM) + pseudo-section
│       ├── helpers.ts       # formatDate, getItemTitle, getProjectTitle, getSectionColor
│       ├── languages.ts     # ISO 639-2/3 → English name mapping
│       ├── urls.ts          # Cross-linking URL builders
│       ├── urlSelection.ts  # URL query-param sync for selection state
│       ├── search.ts        # Generic text-search filter factory
│       ├── pagination.ts    # Generic pagination utility
│       ├── slugify.ts       # URL-safe slug generator (kept in sync with the precompute pipeline)
│       ├── wisskiUrl.svelte.ts  # WissKI navigate URL lookup (Svelte store)
│       ├── featuredCollectionLoader.ts  # Featured collections card builder
│       ├── collectionsRegistry.ts       # Featured collection metadata registry
│       ├── revealOnScroll.ts            # IntersectionObserver-based reveal actions
│       └── cn.ts            # Classname merging (clsx + tailwind-merge)
├── routes/
│   ├── +page.svelte              # Overview dashboard
│   ├── +layout.svelte            # Global layout (Header, Sidebar, data init)
│   ├── whats-new/                # Recent additions (3 / 6 / 12 months)
│   ├── research-sections/        # 13 sections + per-section EntityDashboard
│   ├── projects/                 # Projects + per-project full-parity dashboard
│   ├── research-items/           # Research items browser; detail = KG + sparkline + similar
│   ├── publications/             # Cluster bibliography (ERef + EPub Bayreuth, deduped)
│   ├── people/                   # People directory + per-person dashboard
│   ├── groups/                   # Research groups + per-group dashboard
│   ├── institutions/             # Institutions + per-institution dashboard
│   ├── collections/              # Featured collections index
│   ├── collections/[slug]/       # Collection detail: masonry / map / timeline
│   ├── genres/                   # Genres + per-genre dashboard
│   ├── languages/                # Languages + per-language dashboard
│   ├── locations/                # Locations + browse map + per-location dashboard
│   ├── resource-types/           # Resource types + per-type dashboard
│   ├── subjects/                 # Subjects & Tags + per-entity dashboard
│   ├── project-explorer/         # Cross-project analytical workspace
│   ├── compare/[type]/           # Generic compare (projects | people | institutions |
│   │                             #   subjects | languages | genres) — uses CompareTabs
│   ├── compare-projects/         # Legacy redirect → /compare/projects
│   ├── network/                  # Network visualisation (5 tabs)
│   ├── semantic-map/             # UMAP embedding scatter with similar-items
│   └── sitemap.xml/              # Prerendered sitemap (static + entity-detail URLs)
├── service-worker.js             # PWA shell + asset cache
└── app.css                       # Global styles and Tailwind v4 config

static/
├── manifest.json                # PWA manifest (name, icons, shortcuts, theme colour)
├── robots.txt
├── icons/                       # PWA icons (192 / 512 + maskable)
├── data/
│   ├── manifest.json              # Per-university collection inventory
│   ├── dev/                       # MongoDB exports + WissKI URL mappings
│   ├── manual/                    # Hand-curated supplements (e.g. projectLinks)
│   ├── geo/                       # Natural Earth 110m country GeoJSON for choropleth
│   ├── entity_dashboards/         # Precomputed per-entity dashboard JSON
│   │   ├── manifest.json            # `{ <dir>: [{id, name, count}, ...] }`
│   │   ├── languages/, subjects/, tags/, people/, institutions/,
│   │   ├── genres/, resource-types/, groups/, locations/,
│   │   └── research-sections/, projects/
│   ├── knowledge_graphs/          # Pre-computed ego graphs per entity type + _meta.json
│   ├── embeddings/                # map.json (UMAP) + similar.json (top-K neighbours)
│   ├── projects_metadata_ubt/     # University of Bayreuth (21 collections)
│   ├── projects_metadata_unilag/  # University of Lagos (8 collections)
│   ├── projects_metadata_ujkz/    # Université Joseph Ki-Zerbo (7 collections)
│   ├── projects_metadata_ufba/    # Federal University of Bahia (1 collection)
│   └── external_metadata/         # External collections (BayGlo2025, ILAM)
└── logos/                         # Partner institution logos used across the UI

Data Sources

MongoDB exports — static/data/dev/

• dev.projectsData.json — 92 projects with PIs, members, dates, research sections, institutions, RDSpace references
• dev.persons.json — 1,394 person records with institutional affiliations
• dev.institutions.json — 492 institution records
• dev.groups.json — 84 research-group records
• dev.collections.json — development-only collection items
• dev.researchSections.json — all 13 research sections (Phase 1 + Phase 2 + External)
• dev.geo.json — country / region / subregion / city geolocations with Wikidata coordinates
• dev.wisski_urls.*.json — pre-computed WissKI navigate URL mappings for every entity type (projects, persons, institutions, items, subjects, tags, countries, regions, cities, genres, groups, languages, research sections, resource types)

Partner-university collections — static/data/projects_metadata_*/

3,975 research items across four partners. Each item carries title, contributors (person / institution / group qualifier with roles), subjects (LCSH), tags, language (ISO 639-2/3), location (country / region / city), dates (created, issued, captured), identifiers, physical description, access conditions, and bitstream / URL references.

External collections — static/data/external_metadata/

Items contributed from outside the cluster's partner universities. Tagged university: "external" so they count toward global totals but can be filtered out of per-university views:

• BayGlo2025 — Bayreuth Global / Bayreuth Postkolonial; affiliated with the University of Bayreuth
• ILAM — International Library of African Music (Rhodes University)

Surfaced across the dashboard as virtual projects (Ext_BayGlo2025, Ext_ILAM) under an External pseudo research section, with a dedicated chip in the global filter panel and a dedicated group in the Project Explorer and Compare selectors. Virtual-project definitions live in src/lib/utils/external.ts.

Entity dashboards — static/data/entity_dashboards/

Pre-computed JSON dumps powering the per-entity detail pages and the /compare/[type] selectors. One file per entity instance, organised by directory:

entity_dashboards/
├── manifest.json                # `{ <dir>: [{id, name, count}, ...] }`
├── languages/{code}.json        # 28 entries
├── subjects/{slug}.json         # 620 entries
├── tags/{slug}.json             # 1,100 entries
├── people/{slug}.json           # 1,174 entries
├── institutions/{slug}.json     # 459 entries
├── genres/{slug}.json           # 129 entries
├── resource-types/{slug}.json   # 11 entries
├── groups/{slug}.json           # 84 entries
├── locations/{slug}.json        # 252 entries
├── research-sections/{slug}.json # 6 entries
└── projects/{id}.json           # 38 entries

Each JSON contains all the chart payloads needed to render its entity's dashboard (timeline, stackedTimeline, types, languages, subjects, wordCloud, contributors, roles, heatmap, chord, coContributors, coSubjects, sankey, sunburst, treemap, subjectTrends, locations, selfLocation, geoFlows, contributorNetwork, affiliationNetwork, collabNetwork, meta, items). Generated by scripts/precompute_entity_dashboards.py; the loader lives at src/lib/utils/loaders/entityDashboardLoader.ts.

Knowledge graphs — static/data/knowledge_graphs/

Structural ego graphs pre-computed by scripts/generate_knowledge_graphs.py — one JSON file per entity organised by type:

knowledge_graphs/
├── _meta.json            # Global community labels + top-PageRank nodes
├── items/                # Research-item ego graphs
├── persons/
├── projects/
├── institutions/
├── subjects/
├── tags/
├── locations/
└── genres/

Each graph combines direct metadata edges (IDF-weighted so distinctive relationships look heavier than ubiquitous ones), latent structural edges (Jaccard on shared neighbourhoods + personalised PageRank for multi-hop relevance), and global analysis results (Louvain community membership, PageRank-based centrality) so the UI can reveal discursive communities and key nodes that local co-occurrence alone would miss.

Embeddings — static/data/embeddings/

• map.json — per-item {id, x, y, lowSignal, title, project, university, typeOfResource} for the 2D UMAP scatter
• similar.json — per-item top-12 cosine-similar neighbours keyed by dre_id (lazy-loaded on first selection on /semantic-map, also consumed by SimilarItemsStrip on /research-items?id=…)
• cache.json — full 768-dim Gemini vectors + SHA-256 hashes used for incremental re-embedding; gitignored

Pipelined by scripts/generate_embeddings.py. See scripts/README.md for details.

Cluster publications — static/data/publications.json

146 publications harvested from two University of Bayreuth EPrints repositories and merged into a single normalised payload:

• ERef Bayreuth (EXC 2052 projekt view) — 127 records; the cluster's primary archive
• EPub Bayreuth (Africa Multiple division view) — 50 records; open-access repository

Three EPrints 3 export endpoints are consumed per source: bulk BibTeX (canonical metadata), RSS (deposit dates → year / quarter buckets), and EP3 XML (abstracts, structured keywords, GND author IDs, languages, related / publisher DOIs, volume editors). Cross-source deduplication runs three tiers — DOI (with author / title sanity guard so book-level DOIs don't merge sibling chapters), ISBN (books and periodicals only), and fuzzy (normalized title + year + first-author surname) — folding the 31 cross-listed records into single entries that carry both eref_url and epub_url. Generated by scripts/fetch_publications.py; the loader lives at src/lib/utils/loaders/publicationsLoader.ts.

Geographic data — static/data/geo/

• world-countries-110m.json — Natural Earth 110m country boundaries (TopoJSON) used by ChoroplethMap. Bundled once with the build; smaller than the 50m / 10m datasets at the cost of coastline detail.

Manual data — static/data/manual/

Hand-curated data not sourced from MongoDB:

• projectLinks.json — supplementary project-to-entity links

Scripts

Python pipelines live in scripts/ and cover MongoDB export, WissKI URL generation, thumbnail fetching, knowledge-graph generation, embedding generation, per-entity dashboard precompute, and data slimming. Setup:

python -m venv .venv
.venv/Scripts/pip install -r scripts/requirements.txt

Key scripts (see scripts/README.md for the full list and inputs / outputs):

• fetch_from_mongodb.py — pull research items + reference data from the WissKI MongoDB
• generate_wisski_urls.py — emit pre-computed dev.wisski_urls.*.json lookups for the WissKI Navigate links
• fetch_thumbnails.py — download remote previewImage URLs as local WebP thumbnails so the first paint doesn't hit the public archive
• fetch_publications.py — harvest BibTeX / RSS / EP3 XML from ERef Bayreuth (projekt view) and EPub Bayreuth (Africa Multiple division view), deduplicate across sources by DOI / ISBN / fuzzy title-author-year, and emit static/data/publications.json
• generate_knowledge_graphs.py — compute the ego-graph JSON consumed by EntityKnowledgeGraph and /network?tab=communities
• generate_embeddings.py — run Gemini Embedding 2 over the corpus, project to 2D with UMAP, emit map.json + similar.json
• precompute_entity_dashboards.py — generate one JSON per entity under static/data/entity_dashboards/, plus the global manifest. Supports --from-local, per-type runs, and --dry-run. The manifest writer merges with the existing file so a single-entity run doesn't wipe the others' entries
• slim_data.py — strip development-only fields before shipping production data

Regenerate everything after a metadata refresh:

.venv/Scripts/python scripts/generate_knowledge_graphs.py
.venv/Scripts/python scripts/generate_embeddings.py --scope missing
.venv/Scripts/python scripts/precompute_entity_dashboards.py --entity all --from-local

Deployment

Deploys automatically to GitHub Pages on push to main:

• Workflow: .github/workflows/deploy.yml — npm ci → npm run build → actions/upload-pages-artifact → actions/deploy-pages
• Adapter: @sveltejs/adapter-static pre-renders every route into build/
• Custom domain: static/CNAME pins the site to amira.africamultiple.uni-bayreuth.de; paths.base stays empty so all asset URLs resolve at the apex
• Live URL: https://amira.africamultiple.uni-bayreuth.de/
• CI: .github/workflows/ci.yml runs npm run format:check, npm run lint, npm run check, and npm run build on every push and PR

Credits

Developed by Frédérick Madore for the Digital Research Environment (DRE) of the Africa Multiple Cluster of Excellence, University of Bayreuth.

License

MIT