Girish Knowledge Graph for Context - LLM Wiki

An AI-powered personal knowledge base with an interactive graph visualisation, Claude-powered chat, and a structured wiki for cybersecurity and agentic AI research.

---

What It Does

• Ingest — drop any PDF, PPTX, CSV, or text file into raw/; run the ingest pipeline to extract, convert, and file knowledge automatically across wiki/ pages.
• Graph View — an interactive Cytoscape.js graph visualises all wiki pages (sources, entities, concepts, analyses) as nodes with cross-reference edges.
• Chat — ask natural-language questions; Claude answers using only the wiki content and cites pages inline.
• Wiki — a structured, LLM-maintained knowledge base with pages for sources, entities (people/orgs/tools), concepts (frameworks/ideas), and analyses.

---

Tech Stack

Layer	Technology
Backend	Python · FastAPI · Uvicorn
AI	Anthropic Claude API (claude-opus-4-5)
Frontend	Cytoscape.js · Vanilla JS
Wiki format	Markdown with YAML frontmatter
Source conversion	scripts/ingest.py

---

Project Structure

knowledgraph/
├── raw/                  # Drop source files here (PDF, PPTX, CSV, TXT…)
│   ├── assets/           # Downloaded images referenced in sources
│   └── *.md              # Auto-converted markdown (generated by ingest.py)
├── wiki/                 # LLM-maintained knowledge base
│   ├── index.md          # Content catalog
│   ├── log.md            # Append-only operation log
│   ├── sources/          # One summary page per ingested source
│   ├── entities/         # Pages for people, orgs, products, tools
│   ├── concepts/         # Pages for ideas, frameworks, methodologies
│   └── analyses/         # Query syntheses and lint reports
├── scripts/
│   └── ingest.py         # Source-to-markdown converter
├── static/
│   └── index.html        # Frontend (graph + chat UI)
├── app.py                # FastAPI server
├── CLAUDE.md             # Schema & workflow instructions for Claude
└── AGENTS.md             # Same schema for non-Claude agents

---

Getting Started

1. Install dependencies

pip install fastapi uvicorn anthropic pyyaml python-multipart

2. Set your API key

Create a .env file in the project root:

ANTHROPIC_API_KEY=sk-ant-...

3. Run the app

python app.py

Open http://localhost:8000.

---

API Endpoints

Method	Path	Description
GET	/	Frontend UI
GET	/api/graph	All wiki nodes + edges as JSON
GET	/api/page?path=<rel>	Raw markdown of a single wiki page
GET	/api/chat?q=<question>	Streaming SSE chat (Claude + wiki context)
POST	/api/ingest	Trigger ingest pipeline on a file in raw/

---

Wiki Operations

The wiki is maintained by Claude following the schema in CLAUDE.md. Three operations are supported:

• ingest <filename> — process a new source file and update all relevant wiki pages
• query — ask a question; answer is synthesised from wiki pages with citations
• lint — audit the wiki for contradictions, orphan pages, stale claims, and missing links

---

Current Wiki Contents

Type	Count
Sources	1
Entities	5 (OWASP, NIST, MITRE ATLAS, C2PA, Arup)
Concepts	6 (Prompt Injection, Agentic AI Security, NHI, RAG Security, Digital Provenance, Post-Quantum Cryptography)
Analyses	0

---

License

MIT