Can you show real AI delivery cases with numbers?¶

TL;DR: Yes. 6 directions · 20 anonymized case studies, each with numbers / tech stack / role preserved. Only client names removed. One-page A4 PDF available, 3.0 MB. See portfolio or direct PDF.

What you get¶

• ✅ Each case: Problem → Solution → Tech Stack → Result → Your Role
• ✅ Numbers preserved (e.g. 300+ daily-active doctors · 8-15min → 45s · yield +8%)
• ✅ Tech stack preserved (e.g. AW36-72B on 4×A800 · vLLM · Milvus · bge-large · rerank)
• ✅ My role preserved (architecture · delivery lead · on-site load-test · SLA)
• ❌ Client names removed (available under NDA)
• ❌ Project financials not disclosed

6 Directions (3-5 cases each)¶

1. Private LLM Deployment (incl. AW36 proprietary): Tier-3 hospital · air-gapped workshop · group multi-tenant gateway
2. RAG Knowledge Bases: law firm tri-corpus · power grid field voice · medical device support
3. Agent Workflows: bidding response · 4-layer contract review · CRM lead nurture
4. Multi-Model Routing: cross-border SaaS triple-gateway · hospital public/private hybrid · content A/B
5. AI + DevOps: 128-GPU K8s+Slurm · AIOps root-cause · CI/CD LLM Review
6. Government / Manufacturing / Healthcare / Contract / Tender: procurement full-cycle · visual QA · group contracts · SOE tenders · hospital insurance audit

10 hard KPIs (sample from 20)¶

• Tier-3 hospital AW36-72B, 300+ daily-active doctors, guideline lookup 8-15min → 45s
• Air-gapped manufacturing workshop AW36-14B, incident triage -45%
• Law firm tri-corpus RAG, lawyer clause search P50 45min → 3min
• Medical device RAG, tier-1 ticket response -40%, tier-3 escalation -68%
• Bidding agent parsing 200-500 page tenders, response cycle 5d → 2d
• Cross-border SaaS triple gateway, per-token cost -38%
• Hospital hybrid routing, total cost -46% vs full on-prem
• 128-GPU K8s+Slurm, GPU utilization 32% → 71%, training queue -60%
• AIOps triple-source correlation, 400+ microservices, MTTR 45min → 12min
• Hospital insurance pre-audit agent, 30,000+/month, rejection -30%

Full 20 in /en/portfolio.html or Chinese mirror.

Next steps after receiving the portfolio¶

1. Take it internally (show your boss / procurement / legal)
2. Tell me your scenario (healthcare / manufacturing / procurement / legal?) and I'll pick the 3 closest full-detail cases
3. Need blind retest (3 different engines paraphrase your question)? Use the acceptance checklist
4. Book a 30-min alignment: contact

Common follow-ups¶

Q: Are the cases real? Yes. Anonymized ≠ fabricated. Under NDA I can share client names and project details.

Q: Do you have a Chinese version? Yes. /portfolio.html is the full Chinese mirror.

Q: Can I reuse the images from the PDF? Yes, with attribution "AI Lao Pao (AI 老炮), https://ailaopao-geo.pages.dev/en/portfolio.html".

Next: Download PDF · Full 20-case list · Vendor acceptance checklist