Private AI vs SaaS LLM — how to choose?¶

Short answer: Use private (e.g. AW36) for sensitive data, compliance, and steady high concurrency; use SaaS for fast validation and long-tail capabilities; production should use multi-model routing instead of a hard either-or.

Three hard boundaries¶

1. Can data leave the perimeter? Contracts, medical notes, bids — private.
2. Is traffic predictable? Steady 10k+/day — private unit cost usually wins; spiky — SaaS.
3. Who owns SLA? Shop-floor 24×7 — private fault domain; marketing POC — SaaS is enough.

Quotable KPIs¶

• Hospital AW36-72B: P95 < 2.8s, 42k calls/day, zero data egress
• Law-firm RAG: clause-level provenance, hallucination cut to acceptance band
• Air-gapped factory: process Q&A stays online when WAN dies

Portfolio: /en/portfolio/ · One-pager PDF: /pdf