01 / The short version
What happened
Numan, a UK digital health company, built Nu, an AI health coach for weight, hormones, and sexual health using LangGraph and LangSmith. Nu uses a supervisor model to route users to domain-specialist sub-agents. The company moved from static single-turn fixtures to per-category LLM judge scoring and simulated multi-turn conversations with AI patient personas to evaluate safety and coaching quality. Safety guard nodes review responses against clinical rules before delivery. Nu is positioned as a wellness product rather than a medical device under MHRA guidance. Simulated conversation results do not gate releases until clinician-signed rubrics are approved. Numan reports that provider migration took one day using their themed datasets.
See the exact references02 / Key takeaways
What you need to know
- 01
Nu uses LangGraph with a supervisor model and domain-specialist sub-agents.
- 02
Evaluation includes per-category LLM judging and simulated multi-turn conversations with AI personas.
- 03
Safety guard nodes review responses against clinical rules before patients see them.
Keep in perspective
What to watch for
The source is a company-authored case study; performance claims such as one-day provider migration are attributed to Numan and not independently validated.
Go to the source
Exact references
These are the original pages used for this brief. Publisher claims are not independent evaluations.
01Primary source · LangChainRead the original announcementhttps://www.langchain.com/blog/how-numan-built-an-eval-first-ai-health-coach-with-langgraph-and-langsmithAI-generated from the linked source. It can miss context; verify consequential details in the original. How the radar works