An interactive argument map of ideas: Oliver C. Hirst’s research core, surrounding literature, and machine-inferred conceptual bridges.
This is not a static bibliography dressed up as a graph. It is a live, inspectable system for asking better questions: What connects to what? Which claims carry structural weight? Where are the weak ties, the hidden paths, the conceptual fault lines?
Built as a fast, static, browser-first app with D3.js, modular ES modules, and hard-earned mobile UX refinements.
- Graph exploration with teeth
- Zoom/pan/drag via D3 force simulation
- Hover tooltips for node/link semantics
- Click-to-focus and detail inspection without context loss
- Ranking views that matter
- Overall, EQBSL, PageRank, Degree, Betweenness, Closeness
- AI assistant that acts on the graph, not around it
- Natural-language graph actions (highlight, path finding, focus, inferred edges)
- Free-model discovery from OpenRouter’s models API
- Live graph enrichment
- Add papers from DOI / arXiv / Zenodo input
- Pull adjacent research from Semantic Scholar
- Command palette (
Cmd/Ctrl + K)- Action launcher + graph-aware search
- Mobile UX that behaves like it belongs in 2026
- Bottom-sheet panels, drag handles, safe-area awareness
- scroll-lock, touch targets, and keyboard-aware chat behavior
- Theme support
- Dark + light modes with persisted preference
- Frontend: Vanilla JavaScript (ES modules)
- Visualization: D3.js (
7.8.5, CDN) - Styling: Modular CSS (
base.css,graph.css,panels.css) - Dev server: BrowserSync
- Deployment: Any static host (Netlify/Vercel/GitHub Pages/S3)
npm installnpm run devDefault local URL is printed by BrowserSync (typically http://localhost:3000 or next available port).
npm run startCore config lives in src/config.js.
Important values:
OPENROUTER_KEY— OpenRouter API key used by chat/classification/edge synthesisDEFAULT_MODEL— fallback LLM model- cluster colors, link styles, edge weights (EQBSL/PageRank)
Current implementation uses a client-side key constant in src/config.js. If you rotate keys, update that value and redeploy.
If you are deploying publicly, treat key exposure as inevitable and plan rotation accordingly.
From package.json:
npm run dev— BrowserSync server + live CSS/JS/HTML reloadnpm run start/npm run serve— static serving viaserve
.
├── index.html
├── styles/
│ ├── base.css
│ ├── graph.css
│ └── panels.css
└── src/
├── main.js
├── config.js
├── ai/
│ ├── client.js
│ ├── classify.js
│ ├── edges.js
│ ├── graph-llm.js
│ └── synthesis.js
├── data/
│ ├── arxiv.js
│ ├── cited.js
│ ├── links.js
│ ├── oli.js
│ └── zenodo.js
├── graph/
│ ├── interaction.js
│ ├── render.js
│ └── simulation.js
├── metrics/
│ ├── advanced.js
│ ├── centrality.js
│ ├── community.js
│ └── eqbsl.js
├── panels/
│ ├── chat.js
│ ├── detail.js
│ └── rankings.js
├── sources/
│ ├── fetch.js
│ └── semantic-scholar.js
└── ui/
└── bottom-sheet.js
- Pan/zoom: drag canvas, scroll/pinch to zoom
- Node details: click a node
- Open paper: double-click node (or use detail panel link)
- Filters: top pills for scope/domain/source slicing
- Chat: open
💬 Chatand issue graph-native prompts - Rankings: open
📊 Ranks, switch metric tabs, compare structures - Theme: toggle with
🌙 / ☀️ - Command palette:
Cmd/Ctrl + K
This is a static app. Build step is not required.
- Push repository
- Point static host at repo root
- Publish
index.html+styles/+src/
If you can host static files, you can ship this.
- Move API key out of client bundle
- Add snapshot/export for graph state
- Add richer graph legends and edge provenance inspector
- Add tests for panel state transitions and mobile interactions
Because most literature maps are decorative. This one is operational.
It lets you interrogate a body of work instead of merely admiring it.
No license file is currently included. Add one if you intend public reuse.