Technical perspectives on building AI products
How we architect generative AI for health, clinical practice, and professional workflows — where trust matters.
The peer review ritual must move — from code to ownership
When AI generates, reviews, and maintains the code, the peer-review ritual loses its object. What still matters: qualifying what should be generated, validating properties, and re-appropriating the product.
Why we corroborate generative AI against official databases
Generative AI is probabilistic. In nutrition, a confident wrong estimate can mislead a diabetic. Here is how Zitounix pairs every AI meal analysis with an independent check against 11 official food-composition databases.
Multi-provider architecture: never depend on a single LLM
Why Uptech runs on Alibaba Qwen, Google Gemini, Anthropic Claude, OpenAI, and Ollama — switchable, with automatic fallback. Single-vendor lock-in is a business risk, not just a technical one.
The sourced clinical scribe: AI that doesn't fabricate
Vitary's AI assistant never becomes a source of clinical truth. Every AI-proposed block is tied to an explicit source already in the workspace, with mandatory citations. If the evidence is insufficient, it refuses.
Piloting a project by state, not by conversation
Today's AI coding assistants are chat-based: when the conversation ends, the context is gone. Amar Studio drives an entire project lifecycle through persistent, versioned state — a new agent reads the state and continues, no re-explaining needed.