Knowledge Nexus — Cross-Domain Knowledge Association Engine

English | 中文

Discover deep connections between seemingly unrelated concepts across domains. Weave isolated knowledge into an interconnected intelligence network, powered by AI.

---

🎯 Vision

Human knowledge is scattered across disciplines, yet many concepts are deeply connected:

Domain A	↔	Domain B
Biology: Natural Selection	↔	CS: Genetic Algorithm
Physics: Annealing	↔	Optimization: Simulated Annealing
Neuroscience: Neural Networks	↔	Deep Learning: Artificial Neural Networks
Economics: Game Theory	↔	Multi-Agent: Reinforcement Learning

Knowledge Nexus aims to:

1. Build intra-domain knowledge graphs — Map citation, inheritance, and improvement relationships between SOTA works within a domain
2. Discover cross-domain associations — Use AI to identify structural similarities and concept transfers across fields
3. Generate new knowledge hypotheses — Infer potential cross-domain inspirations and innovation directions

✨ Features

📚 Knowledge Base Management

• 8 knowledge node types: phenomenon, theorem, law, method, concept, principle, process, structure
• 19 domains covering natural sciences, social sciences, humanities, and engineering
• Paper metadata with PDF storage, DOI, arXiv ID, citation counts, impact scoring, and LLM-generated summaries
• Batch import via scripts or crawler, batch delete with cascading relation cleanup

🕸️ Interactive Knowledge Graph

• Cytoscape.js-powered graph visualization with domain-colored nodes
• 5 layout algorithms: Force-directed (Cose), Circle, Concentric, Breadthfirst, Grid
• Search by keyword, filter by domain, toggle cross-domain mode
• Zoom, pan, drag, and click-to-focus interactions
• Full graph view + subgraph exploration (configurable depth 1–3)
• 13 relation types: CITES, BUILDS_ON, IMPROVES, ANALOGOUS_TO, INSPIRES, INSPIRED_BY, PART_OF, ENABLES, RELATED_TO, CONTRADICTS, COMPETES_WITH, USED_BY, REVIEWS

🤖 AI Discovery Engine (LLM-Powered)

• Cross-Domain Discovery — AI scans knowledge nodes to find 5–10 hidden cross-domain associations per run, with confidence scores and visual indicators
• Pair Analysis — 6-dimensional deep analysis: structural analogy, causal links, complementarity, unified framework, and more
• Knowledge Derivation — Select 2–10 nodes and derive abstract patterns, transfer ideas, missing links, and new hypotheses with feasibility ratings
• 💬 Conversational AI Assistant — ChatGPT-style multi-turn conversation interface for natural language exploration of the knowledge base, with skill-based auto-routing (search, discover, analyze, derive, summarize)
  • One-click copy to clipboard for any AI response
  • Regenerate (retry) button to re-generate any AI response, similar to GPT/Claude/Gemini — discards subsequent messages and retries from that point
  • Structured data cards for search results, discoveries, pair analyses, and derivations rendered inline
• Save discoveries as "pending review" or auto-confirm into the knowledge graph
• Domain filtering to focus discovery on specific fields
• Fuzzy matching with 3-level strategy for robust node identification
• Compatible with any OpenAI-format LLM API (Doubao, DeepSeek, OpenAI, Ollama)

🕷️ Smart Paper Crawler

• Multi-source crawling: OpenAlex, Semantic Scholar, arXiv
• Quality scoring based on citation count, venue prestige, and SOTA records
• Auto-download Open Access PDFs
• Rate-limited, resumable, deduplicated

🔍 Graph Exploration

• Domain-filtered subgraphs with configurable depth (1–3)
• Cross-domain mode highlighting inter-field connections
• Node detail panel with full metadata
• Citation threshold filtering for papers

📐 Architecture

graph TB
    subgraph Frontend["🖥️ Frontend — React + TypeScript"]
        direction LR
        GV["📊 Graph View<br/><i>Cytoscape.js</i>"]
        KM["📚 Knowledge<br/>Management"]
        AI["🤖 AI Discovery<br/><i>3-Tab Panel</i>"]
        CR["🕷️ Crawler<br/>Dashboard"]
    end

    subgraph Backend["⚙️ Backend — FastAPI + Python"]
        direction LR
        PS["📄 Paper<br/>Service"]
        GS["🕸️ Graph<br/>Service"]
        AE["🧠 AI Engine<br/><i>LLM Analyzer</i>"]
        CS["🔍 Crawler<br/>Service"]
    end

    subgraph Storage["💾 Data Layer"]
        direction LR
        DB[("🗄️ SQLite / PostgreSQL<br/><i>Papers · Nodes · Relations</i>")]
        FS["📁 File Storage<br/><i>PDF Papers</i>"]
    end

    subgraph External["🌐 External Services"]
        direction LR
        LLM["🤖 LLM API<br/><i>Doubao · DeepSeek<br/>OpenAI · Ollama</i>"]
        OA["📖 OpenAlex<br/><i>Paper Metadata</i>"]
        SS["🔬 Semantic Scholar<br/><i>Citations</i>"]
    end

    Frontend -->|"REST API<br/>(Vite Proxy)"| Backend
    PS --> DB
    PS --> FS
    GS --> DB
    AE --> LLM
    AE --> DB
    CS --> OA
    CS --> SS
    CS --> DB

    style Frontend fill:#e6f3ff,stroke:#4a90d9,stroke-width:2px
    style Backend fill:#f0f7e6,stroke:#5cb85c,stroke-width:2px
    style Storage fill:#fff8e6,stroke:#f0ad4e,stroke-width:2px
    style External fill:#fce6f0,stroke:#d9534f,stroke-width:2px

Loading

Data Flow

flowchart LR
    A["📥 Import<br/><i>Scripts · Crawler · Manual</i>"] --> B["🗄️ Knowledge Base<br/><i>Nodes + Papers + Relations</i>"]
    B --> C["🕸️ Graph Visualization<br/><i>Cytoscape.js</i>"]
    B --> D["🧠 AI Discovery Engine<br/><i>LLM Analysis</i>"]
    D -->|"New associations<br/>& hypotheses"| B
    D --> E["💡 New Knowledge<br/><i>Cross-domain insights</i>"]

    style A fill:#e8f5e9,stroke:#4caf50
    style B fill:#e3f2fd,stroke:#2196f3
    style C fill:#fff3e0,stroke:#ff9800
    style D fill:#fce4ec,stroke:#e91e63
    style E fill:#f3e5f5,stroke:#9c27b0

Loading

🛠️ Tech Stack

Layer	Technology
Frontend	React 18, TypeScript, Ant Design, Cytoscape.js, Vite
Backend	Python 3.11+, FastAPI, SQLAlchemy, Pydantic
Database	SQLite (dev), PostgreSQL (prod-ready)
AI/LLM	OpenAI-compatible API (Doubao, DeepSeek, OpenAI, Ollama)
Crawler	httpx, OpenAlex API, Semantic Scholar API

🚀 Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• An LLM API key (Doubao, DeepSeek, OpenAI, or local Ollama)

1. Clone the Repository

git clone https://github.com/Harris-H/knowledge-nexus.git
cd knowledge-nexus

2. Backend Setup

cd backend

# Create virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env — set your LLM_API_KEY

# Start the backend
uvicorn app.main:app --host 0.0.0.0 --port 8082 --reload

3. Frontend Setup

cd frontend

# Install dependencies
npm install

# Start dev server (auto-proxies API to backend)
npm run dev

4. Initialize Knowledge Base (Optional)

# Add cross-domain knowledge nodes
cd scripts
python add_cross_domain_knowledge.py
python add_cross_domain_knowledge_v2.py
python add_cs_knowledge_v3.py
python add_speech_ai_knowledge.py
python update_speech_domain.py

5. Open in Browser

Visit http://localhost:3001 (or the port shown in terminal).

📁 Project Structure

knowledge-nexus/
├── README.md                    # English documentation
├── README_zh.md                 # 中文文档
├── backend/                     # FastAPI backend
│   ├── app/
│   │   ├── api/                 # API routes (papers, graph, ai, crawler)
│   │   ├── models/              # SQLAlchemy models (Paper, KnowledgeNode, Relation)
│   │   ├── schemas/             # Pydantic request/response models
│   │   ├── services/            # Business logic
│   │   │   ├── ai/              # LLM-powered discovery engine
│   │   │   ├── crawler/         # Paper crawling service
│   │   │   └── ...
│   │   └── core/                # Config, database setup
│   ├── .env.example             # Environment template
│   └── requirements.txt
├── frontend/                    # React + TypeScript frontend
│   ├── src/
│   │   ├── pages/               # Main pages
│   │   │   ├── GraphPage.tsx    # Knowledge graph visualization
│   │   │   ├── AIDiscoveryPage.tsx  # AI discovery (3 tabs)
│   │   │   ├── PapersPage.tsx   # Paper management
│   │   │   ├── KnowledgeNodesPage.tsx  # Knowledge node management
│   │   │   └── CrawlerPage.tsx  # Paper crawler
│   │   ├── api/                 # API client
│   │   ├── components/          # Shared components
│   │   └── types/               # TypeScript types
│   └── package.json
├── scripts/                     # Data import scripts
├── docs/                        # Design documents
│   ├── tech-stack.md
│   ├── architecture.md
│   ├── api-design.md
│   └── crawler-design.md
├── storage/                     # File storage (PDFs)
└── docker-compose.yml           # Docker setup (optional)

📊 Current Knowledge Base

Domain	Nodes	Papers	Description
💻 Computer Science	49	~26	Full AI stack: Backpropagation → Transformer → LLM → Agent → MCP
🎤 Speech AI	12	18	ASR, TTS, Voice Cloning, Neural Audio Codec, Speech LLM
🧠 Philosophy	22	-	Reductionism, systems thinking, emergence, epistemology
🎨 Art	12	-	Golden ratio, generative art, color theory, Gestalt, montage
🧬 Biology	10	-	Evolution, genetics, CRISPR, symbiosis, central dogma
⚛️ Physics	10	-	Thermodynamics, quantum mechanics, Noether's theorem, superconductivity
📊 Mathematics	10	-	Graph theory, optimization, topology, Gödel's incompleteness
🧪 Psychology	10	-	Conditioning, cognitive dissonance, working memory, conformity
🔬 Chemistry	12	-	Periodic law, acid-base theory, redox, chirality, spectroscopy
🌿 Ecology	12	-	Competitive exclusion, succession, nitrogen cycle, biodiversity
💰 Economics	12	-	Supply-demand, Nash equilibrium, prospect theory, externalities
⚙️ Engineering	12	-	FEA, redundancy, modular design, fatigue failure, PLM
🧠 Neuroscience	12	-	Hebbian learning, LTP, synaptic pruning, BCI, lateral inhibition
👥 Sociology	12	-	Social capital, weak ties, Dunbar's number, labeling theory
🏥 Medicine	12	-	Dose-response, precision medicine, microbiome, medical imaging
🧠 Cognitive Science	12	-	Metacognition, dual process, change blindness, situated cognition
🌱 Life Science	12	-	Cell theory, autophagy, epigenetics, protein folding
⚔️ Military Science	12	-	OODA loop, Lanchester's laws, wargaming, Art of War
📜 History	12	-	Path dependence, Longue durée, Great Divergence, Occam's razor

Total: 267 nodes, 44 papers, 429+ relations, 19 domains, 8 node types

🤖 LLM Configuration

Knowledge Nexus supports any OpenAI-compatible LLM API. Edit backend/.env:

# Doubao (ByteDance) — Default
LLM_API_KEY=your-api-key
LLM_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
LLM_MODEL=doubao-seed-2-0-lite-260215

# DeepSeek
# LLM_BASE_URL=https://api.deepseek.com/v1
# LLM_MODEL=deepseek-chat

# OpenAI
# LLM_BASE_URL=https://api.openai.com/v1
# LLM_MODEL=gpt-4o-mini

# Local Ollama
# LLM_BASE_URL=http://localhost:11434/v1
# LLM_MODEL=qwen2.5

📸 Screenshots

🕸️ Knowledge Graph Visualization

Interactive graph powered by Cytoscape.js — domain-colored nodes, 13 relation types, search & filter toolbar.

🤖 AI Discovery Chat

ChatGPT-style conversational assistant with copy & regenerate buttons, structured data cards, and skill-based auto-routing.

📚 Knowledge Node Management

Browse, filter, and manage 267+ knowledge nodes across 19 domains with type tags, summaries, and year metadata.

🗺️ Roadmap

• Semantic search with vector embeddings
• PDF auto-parsing and metadata extraction
• Multi-user collaboration
• Knowledge timeline view
• Export to standard formats (RDF, OWL)
• Plugin system for custom domain adapters

📄 License

MIT