Nikhil Deshpande

Healthcare · 2024–2026

Luke AI

One expert, eighty practitioners

The problem

Around 80 nutritionists work inside the practice. Not one of them can get time with Luke. A decade of his clinical thinking sat in documents and in his head, and every practitioner was working from whatever part of it they had picked up.

So the question was simple to ask and hard to build. What if Luke’s brain is available to every one of them, to ask against their own client’s case?

The system

Retrieval over the practice’s own corpus, answering in Luke’s terms rather than the internet’s. The same layer serves two faces. The coaches use it inside their work, against a specific client, as reference. Clients meet it as a conversational product.

The retrieval architecture is the method, not the point. The point is that one expert’s judgement became something eighty people could consult at the moment they needed it, inside a business where a wrong answer is a clinical event.

What happened

Shipped to production and used by the team. No clinical safety incident. That constraint was never negotiable, and it shaped every decision about what the system would refuse to answer.