Live Demo — Healthcare Knowledge Graph

200 clinical concepts. 500+ relationships.
Compressed to 12KB of portable markdown.

This is what a healthcare knowledge graph looks like when it's structured for AI reasoning, compressed to .md, and ready to deploy into any environment — Claude, ChatGPT, Cursor, MCP, or raw API.

Raw Data
~2MB
Knowledge Graph
200 nodes
Graph Distillation
500+ edges
Compressed .md
12KB
Token Density
170x
170x
Token Density
93%
Token Compression
12KB
.md File Size
~3K
Tokens (vs ~500K raw)
0.002
kg CO₂ per query

Interactive Healthcare Knowledge Graph

What the AI actually receives — structured, compressed, portable

healthcare-cardiology.md

# Cardiology Domain Knowledge Graph ## Entities ### Conditions - Atrial Fibrillation | ICD: I48 | prevalence: 2.7M US - Heart Failure | ICD: I50 | prevalence: 6.2M US - subtypes: HFrEF (EF≤40%), HFpEF (EF≥50%), HFmrEF - Coronary Artery Disease | ICD: I25 | #1 killer US - Hypertension | ICD: I10 | prevalence: 116M US - Deep Vein Thrombosis | ICD: I82 ### Medications - Metoprolol | class: beta-blocker | route: PO - Lisinopril | class: ACE-inhibitor | route: PO - Warfarin | class: anticoagulant | INR target: 2-3 - Apixaban | class: DOAC | no INR monitoring - Amiodarone | class: antiarrhythmic | ⚠️ toxicity ### Diagnostics - Echocardiogram | CPT: 93306 | measures: EF, valves - Cardiac Catheterization | CPT: 93458 | gold standard - BNP/NT-proBNP | CPT: 83880 | HF biomarker - Troponin | CPT: 84484 | MI biomarker ## Relationships AFib → TREATED_BY → Apixaban (first-line DOAC) AFib → TREATED_BY → Warfarin (if valve disease) AFib → RISK_FACTOR_FOR → Stroke (5x risk) AFib → DIAGNOSED_BY → ECG + Echo HFrEF → TREATED_BY → Metoprolol (mortality ↓35%) HFrEF → TREATED_BY → Lisinopril (afterload ↓) HFrEF → MONITORED_BY → BNP (>400 = severe) CAD → DIAGNOSED_BY → Cardiac Cath CAD → RISK_FACTOR → Hypertension Hypertension → TREATED_BY → Lisinopril Warfarin → INTERACTS_WITH → Amiodarone ⚠️ ↳ RULE: ↑INR 50-70%. Reduce warfarin dose 30-50%. Metoprolol → CONTRAINDICATED → Bradycardia ↳ RULE: Hold if HR < 60 bpm Apixaban → REQUIRES → CrCl assessment ↳ RULE: Reduce dose if CrCl 15-29, avoid if <15 ## Traversal Examples Q: Patient with AFib + CKD Stage 4. Anticoagulation? AFib → TREATED_BY → Apixaban Apixaban → REQUIRES → CrCl CKD Stage 4 → CrCl 15-29 → DOSE_ADJUST Apixaban → Answer: Apixaban 2.5mg BID (reduced dose) Q: Patient on Warfarin, starting Amiodarone? Warfarin → INTERACTS → Amiodarone → RULE: Reduce warfarin 30-50%. Check INR in 3-5 days. → Answer: Cut warfarin dose, weekly INR x4 weeks

Why This Works

The raw data — clinical guidelines, drug databases, ICD codes, CPT codes, formularies — is ~2MB of unstructured text across dozens of sources.

The knowledge graph extracts 200 entities and 500+ typed relationships. Conditions connect to treatments connect to contraindications connect to diagnostics connect to billing codes. Every connection is traversable.

The .md compression distills the graph into ~12KB of structured markdown. That's ~3,000 tokens. The raw data would cost ~500,000 tokens. 170x more efficient.

The carbon impact — fewer tokens = less compute = lower energy. A single query against the compressed .md uses ~0.002 kg CO₂ vs ~0.34 kg for the raw corpus. 99.4% reduction in carbon per query.

The portability — this .md file works everywhere. Drop it into Claude as a project file. Load it into ChatGPT. Reference it in Cursor. Serve it via MCP. Email it. It's just text. But it's structured text that makes any model reason like a specialist.

The traversal — the model doesn't just retrieve text chunks. It follows relationship chains: AFib → Apixaban → CrCl → CKD Stage 4 → dose adjustment. Multi-hop reasoning. Auditable. Explainable. No hallucination.

One .md file. Every AI environment.

The same compressed knowledge graph works across every AI tool — no reformatting, no API integration, no vendor lock-in.

Claude Projects
Upload as project knowledge
ChatGPT
Custom GPT instructions
Cursor / Windsurf
.cursor/rules or context file
Claude Code
CLAUDE.md project context
MCP Server
Serve as tool context
API / SDK
System prompt injection
RAG Pipeline
Pre-retrieval context layer
Email / Slack
Just paste it. It's text.

Want this for your domain?

Pick any vertical. 30 minutes. I'll show you what your knowledge graph looks like compressed to .md.

Book a Demo