English · Türkçe · Français · 中文 · العربية · Русский
132 terms · 531 relationships · 11 categories · 6 languages
The AI revolution has introduced hundreds of new technical terms into our vocabulary in just the last few years. From transformer architectures to prompt injection attacks, from quantization formats to agentic AI frameworks — the terminology landscape is expanding faster than any textbook, course, or documentation can keep up with.
This is a real problem across every technical role:
- A developer integrating an LLM API encounters terms like KV-cache, temperature, top-p, system prompt — but how do they connect? What are the security implications of each?
- A cybersecurity professional auditing an AI system needs to understand indirect prompt injection, RAG poisoning, model denial-of-service — but where is the map?
- A DevSecOps engineer deploying vLLM on Kubernetes needs to know about tensor parallelism, continuous batching, PagedAttention, CUDA stream hijacking — all in context, all connected.
- A researcher needs to see how differential privacy relates to training data extraction, how RLHF connects to alignment, how MoE changes the inference landscape.
Existing resources are scattered. Blog posts cover individual topics. Documentation lives in silos. No single resource maps the connections between AI/ML concepts, their security implications, and practical configurations — across multiple languages.
This project fills that gap. An interactive, force-directed knowledge graph that visualizes how 132 AI/ML concepts relate to each other, with deep-dive content on security vulnerabilities, hardening configurations, research papers, and actionable tips — in 6 languages.
Every node in the graph opens a rich, structured knowledge card:
| Section | Content | Example (RAG) |
|---|---|---|
| Description | 2-3 paragraph explanation with inline links | How retrieval-augmented generation works, key components, why it reduces hallucination |
| Analogy | Real-world comparison for intuition | "Like an open-book exam — the model looks up answers instead of memorizing everything" |
| Security | Mechanism-specific vulnerabilities with severity | Document poisoning (high), retrieval manipulation (high), context injection (medium) |
| Configuration | Working code examples for hardening | Python: chunk size limits, embedding validation, retrieval filtering |
| Research | Papers, official docs, OWASP guides | arXiv links, vendor documentation, security advisories |
| Tips | Quick actionable recommendations | "Always validate retrieved documents before injecting into the prompt" |
Every term's security section is tied to its own mechanism — not generic advice.
| Term | Security Focus |
|---|---|
| GPU | Memory isolation vulnerabilities, side-channel attacks, CUDA stream hijacking |
| Transformer | Self-attention memorization, positional encoding manipulation |
| KV-Cache | Cross-tenant cache poisoning, attention state leakage |
| Tokenizer | Homoglyph attacks, BPE token smuggling, Unicode manipulation |
| RAG | Document poisoning, retrieval manipulation, context injection |
| LoRA | Adapter poisoning, malicious adapter injection on shared hubs |
| vLLM | Unauthorized API access, model theft via serving endpoints |
| HuggingFace | Pickle deserialization RCE, model supply chain attacks |
| Prompt Injection | RAG-based attacks, tool-use hijacking, agent manipulation |
| Chat Template | Tokenizer exploitation, system prompt boundary bypass |
The graph uses an Apple-inspired minimal visual language called Snow:
- Clean circles with low-opacity category fill and a subtle center dot
- Quadratic bezier curves for relationship links — nearly invisible, illuminated on interaction
Interfont family with Apple HIG spacing and weight hierarchy- Light mode: pure white
#fffbackground,#f5f5f7surface cards - Dark mode:
#1c1c1ebackground,#2c2c2esurfaces — native Apple dark palette - No gradients, no glow effects, no animations on nodes — pure clarity
| Feature | Details |
|---|---|
| Theme | Light & Dark with smooth transition |
| Node sizing | Proportional to connection count |
| Link behavior | Transparent by default, highlighted on hover/select |
| Persistent highlight | Selected node keeps connections visible while panel is open |
| Spotlight search | Cmd+K / Ctrl+K to search across all terms |
| Responsive | Desktop, tablet, and mobile layouts |
| Category | Color | Example Terms |
|---|---|---|
| Hardware | #f87171 |
GPU, CUDA, VRAM, TPU, DGX Spark |
| Architecture | #60a5fa |
Transformer, Attention, Encoder-Decoder, MoE |
| Training | #a78bfa |
Pretraining, Fine-tuning, RLHF, LoRA, DPO |
| Quantization | #34d399 |
FP16, INT4, GPTQ, AWQ, GGUF |
| Inference | #fbbf24 |
vLLM, KV-Cache, PagedAttention, Speculative Decoding |
| Prompting | #f472b6 |
System Prompt, RAG, Few-Shot, Chain-of-Thought |
| Tools & Agents | #a3e635 |
MCP, Tool Use, Function Calling, Agent, Agentic AI |
| Model Types | #22d3ee |
LLM, VLM, Foundation Model, SLM |
| AI Security | #ef4444 |
Prompt Injection, Jailbreak, Data Poisoning, AI Firewall |
| Fundamentals | #8b5cf6 |
Neural Network, Deep Learning, Gradient, Backpropagation |
| Ecosystem | #f59e0b |
HuggingFace, Ollama, LangChain, Gradio |
| Language | Code | Direction |
|---|---|---|
| English | en |
LTR |
| Turkish | tr |
LTR |
| French | fr |
LTR |
| Chinese | zh |
LTR |
| Arabic | ar |
RTL |
| Russian | ru |
LTR |
All descriptions, analogies, security findings, tips, and relationship labels are available in all 6 languages. The UI switches dynamically, including full RTL support for Arabic.
- Bun v1.0+
git clone https://github.com/badchars/ai-knowledge-graph.git
cd ai-knowledge-graph
bun install
bun run buildbun run dev # builds and opens in browserbun run validate # checks all YAML files against schemaThe repo includes CI/CD workflows that handle everything:
- Fork or push this repo to your GitHub account
- Go to Settings > Pages
- Under "Build and deployment", select Source: GitHub Actions
- Done — every push to
maintriggers build & deploy
push to main → bun install → build → upload dist/ → deploy to Pages
bun run build
# Upload dist/ to any static host:
# GitHub Pages, Netlify, Vercel, Cloudflare Pages, etc.Every pull request that touches data/** automatically runs:
bun run validate— YAML schema checkbun run build— ensures the site builds- If either fails, the PR cannot merge
| Action | How |
|---|---|
| Open term details | Click any node |
| See connections | Hover over a node |
| Search | Cmd+K or Ctrl+K or click the search bar |
| Zoom | Scroll wheel |
| Pan | Click and drag the background |
| Rearrange | Drag individual nodes |
| Switch language | Language selector in the header |
| Toggle theme | Sun/moon icon in the header |
| Close panel | ESC key |
| Navigate related terms | Click related terms in the detail panel |
We welcome contributions from anyone — AI researchers, security professionals, developers, students.
# 1. Fork and clone
git clone https://github.com/YOUR_USERNAME/ai-knowledge-graph.git
cd ai-knowledge-graph && bun install
# 2. Create a new term
cp data/terms/_template.yaml data/terms/my-new-term.yaml
# 3. Edit the YAML file
# 4. Add relationships in data/relationships.yaml
# 5. Validate and test
bun run validate && bun run build
# 6. Submit a PR| Type | Difficulty | Impact |
|---|---|---|
| Add a new term | Medium | High |
| Add translations | Easy | High |
| Fix descriptions | Easy | Medium |
| Add security findings | Medium | Very High |
| Add research links | Easy | Medium |
| Add config examples | Medium | High |
| Report inaccuracies | Easy | High |
Each term is a standalone YAML file in data/terms/:
id: my-new-term # must match filename
name: MyTerm # short display name
fullName: My Full Term Name # expanded name
category: architecture # one of 11 categories
desc:
en: |
English description with [inline links](https://example.com).
Multiple paragraphs separated by blank lines.
tr: |
Turkish description...
fr: ""
zh: ""
ar: ""
ru: ""
analogy:
en: "A real-world comparison..."
security:
- title:
en: Vulnerability Name
description:
en: How this vulnerability works.
severity: high # critical | high | medium | low
research:
- url: https://arxiv.org/abs/xxxx.xxxxx
label:
en: Paper Title
tips:
- en: A quick actionable tip# data/relationships.yaml
- source: my-new-term
target: transformer
label:
en: uses
tr: kullanir
fr: utilise
zh: 使用
ar: يستخدم
ru: используетNo code required — use our forms:
See CONTRIBUTING.md for the full guide.
ai-knowledge-graph/
├── build.ts # YAML → static site generator
├── package.json # build, dev, validate scripts
├── CONTRIBUTING.md
├── LICENSE # MIT
│
├── data/
│ ├── terms/ # 132 individual YAML files
│ │ ├── _template.yaml # contributor template
│ │ ├── gpu.yaml
│ │ ├── transformer.yaml
│ │ ├── prompt-injection.yaml
│ │ └── ...
│ ├── relationships.yaml # 531 edges (6-language labels)
│ ├── categories.yaml # 11 categories with colors
│ └── ui.json # UI strings (6 languages)
│
├── public/
│ ├── css/style.css # Snow theme (light + dark)
│ └── js/
│ ├── graph.js # D3.js force simulation
│ ├── i18n.js # Language switching
│ └── theme.js # Light/dark toggle
│
├── templates/
│ ├── index.html # Graph page template
│ └── term.html # Term detail page template
│
├── scripts/
│ ├── validate.ts # YAML schema validator
│ └── migrate-to-yaml.ts # One-time JSON → YAML migration
│
└── .github/
├── workflows/
│ ├── validate.yml # PR validation
│ └── deploy.yml # GitHub Pages deploy
└── ISSUE_TEMPLATE/
├── new-term.yml
├── update-term.yml
└── add-translation.yml
┌─────────────────────────────────────────────────────────┐
│ Data Layer (YAML) │
│ │
│ data/terms/*.yaml 132 individual term files │
│ data/relationships 531 edges with 6-language labels │
│ data/categories 11 categories with colors │
│ data/ui.json UI strings for 6 languages │
└────────────────────────────┬────────────────────────────┘
│
bun run build
│
┌────────────────────────────▼────────────────────────────┐
│ Build Output (dist/) │
│ │
│ index.html Graph page (data inlined) │
│ term/*.html 132 individual term pages │
│ data/graph.json Aggregated data for API use │
│ css/, js/ Static assets │
└────────────────────────────┬────────────────────────────┘
│
GitHub Actions
│
┌────────────────────────────▼────────────────────────────┐
│ GitHub Pages │
│ │
│ Static hosting — zero backend, zero database │
│ Auto-deploys on every push to main │
└─────────────────────────────────────────────────────────┘
| Role | How This Helps |
|---|---|
| Developers | Understand AI/ML terms you encounter daily — what they mean, how they connect, security implications |
| Security Professionals | Map the AI attack surface — prompt injection to model theft, with mechanism-specific vulnerabilities |
| DevSecOps Engineers | Hardening guides with real code — every security finding includes config examples |
| Pentesters | Architecture knowledge reveals attack vectors — understand AI systems deeply enough to test them |
| AI/ML Engineers | Quick reference for the ecosystem — tools, frameworks, and techniques in context |
| Students & Researchers | Analogies make concepts intuitive, research links point to primary sources |
| Component | Technology |
|---|---|
| Runtime | Bun |
| Visualization | D3.js v7 |
| Data | YAML (individual files per term) |
| Build dependency | js-yaml |
| Hosting | GitHub Pages |
| Frontend | Vanilla HTML / CSS / JS |
| CI/CD | GitHub Actions |
- Live deployment on GitHub Pages
- Fuzzy search across all term content
- Side-by-side term comparison
- Export to PDF / Markdown
- Embeddable widget for blogs and docs
- API endpoint for programmatic access
- Community voting on term accuracy
- Auto-generated terms from OWASP / MITRE databases
- Japanese and Korean language support