Healthcare AI GEO¶

Patient / doctor / hospital procurement now start with AI. If you're not cited, you're not considered.

How healthcare customers use AI (3 journeys)¶

1. Patient B2C¶

Actual prompts (ChatGPT / Perplexity / Yuanbao): - "best XX hospital in Shanghai for X condition" - "second-opinion review of my MRI report" - "orthopedic surgeon recommendations, Beijing, English-speaking"

AI cites: hospital website / doctor profile / research paper / patient forum. Reality: patients now filter shortlists via AI before making appointments.

2. Doctor / Specialist¶

Physicians use AI to look up: guidelines, drug interactions, rare presentations, discharge summary templates. Your professional publications (case reviews, guideline commentary, protocol templates) = potential citation → doctor trust.

3. Hospital B2B procurement (medical device / EMR / AI vendor)¶

Hospital IT / medical affairs / procurement use AI to shortlist vendors. Your public compliance evidence + Level-3 protection assessment + real deployment KPI = potential citation.

5 hard PHI red lines (non-negotiable)¶

1. PHI stays on intramural network (unless hospital compliance explicitly approves public egress)
2. Full audit log (who / when / query / answer), retained ≥ 180 days
3. Prompt + model + retrieval index all versioned + rollback-able
4. Intramural IT + medical affairs + procurement 3-way sign-off — vendor cannot unilaterally push
5. Level-3 protection / GDPR / HIPAA compliance assessment front-loaded before sizing

Violate any = no purchase. Regulator can shut down.

Healthcare AI delivery cases¶

Case 01 · Tier-3 Hospital AW36-72B Private Deployment + Doctor Assistant¶

• Guideline lookup 8-15 min → 45s · 300+ daily-active doctors · discharge-summary +50% · Level-3 protection passed
• AW36-72B on 4×A800, vLLM, Milvus, bge-large, BGE-reranker
• Full detail: Case 01

Case 11 · Hospital Public/Private Hybrid Routing¶

• Total cost -46% vs full on-prem · zero PHI egress
• Sensitivity-tier routing (PHI internal, non-PHI public)
• Full detail: Case 11

Case 20 · Hospital Insurance Pre-Audit Agent¶

• 30,000+/month claims pre-audited · rejection rate -30%
• National insurance rules + intramural clinical norms fusion
• Full detail: Case 20

Case 06 · Medical Device Support RAG¶

• Tier-1 response -40% · tier-3 escalation -68% · CSAT +12%
• bge-large + BGE-reranker, versioned product docs, feedback-loop retrain
• Full detail: Case 06

What healthcare projects won't do¶

• ❌ Cloud GPT-4 on PHI — hard no
• ❌ AI direct decision on treatment / prescription — AI outputs suggestion, doctor decides
• ❌ No provenance answers — regulatory audit fails
• ❌ Skip compliance assessment to hit go-live — every "faster launch" story becomes an incident

Cost range¶

200k - 3M CNY depending on scope.

• Single-department RAG (doctor assistant, single specialty): 200k - 500k
• Multi-department + insurance / operations agents: 800k - 2M
• Full hospital AI platform + integration: 2M - 3M+

Related¶

• Direction · Private LLM Deployment
• Direction · RAG Knowledge Base
• How-To · Private AI Sizing
• How-To · RAG Cold Start
• Buyer Q · Hospital private-LLM cases
• 中文详细版

Next: Portfolio PDF · 30-min alignment